kar/stt-whisper.cpp-cpu-api-py
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

init

Browse Source
This commit is contained in:
Kar
2025-06-17 15:53:01 +05:30
commit 4d20931ecc
411 changed files with 180695 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperContext.java Normal file
View File

@@ -0,0 +1,41 @@
package io.github.ggerganov.whispercpp;
import com.sun.jna.Structure;
import com.sun.jna.ptr.PointerByReference;
import io.github.ggerganov.whispercpp.ggml.GgmlType;
import io.github.ggerganov.whispercpp.WhisperModel;
import io.github.ggerganov.whispercpp.params.WhisperContextParams;
import java.util.List;
public class WhisperContext extends Structure {
int t_load_us = 0;
int t_start_us = 0;
/** weight type (FP32 / FP16 / QX) */
GgmlType wtype = GgmlType.GGML_TYPE_F16;
/** intermediate type (FP32 or FP16) */
GgmlType itype = GgmlType.GGML_TYPE_F16;
// WhisperModel model;
public PointerByReference model;
// whisper_vocab vocab;
// whisper_state * state = nullptr;
public PointerByReference vocab;
public PointerByReference state;
/** populated by whisper_init_from_file_with_params() */
String path_model;
WhisperContextParams params;
// public static class ByReference extends WhisperContext implements Structure.ByReference {
// }
//
// public static class ByValue extends WhisperContext implements Structure.ByValue {
// }
//
// @Override
// protected List<String> getFieldOrder() {
// return List.of("t_load_us", "t_start_us", "wtype", "itype", "model", "vocab", "state", "path_model");
// }
}

207
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java Normal file
View File

@@ -0,0 +1,207 @@
package io.github.ggerganov.whispercpp;
import com.sun.jna.Native;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.bean.WhisperSegment;
import io.github.ggerganov.whispercpp.params.WhisperContextParams;
import io.github.ggerganov.whispercpp.params.WhisperFullParams;
import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
import java.io.File;
import java.io.FileNotFoundException;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
/**
* Before calling most methods, you must call `initContext(modelPath)` to initialise the `ctx` Pointer.
*/
public class WhisperCpp implements AutoCloseable {
private WhisperCppJnaLibrary lib = WhisperCppJnaLibrary.instance;
private Pointer ctx = null;
private Pointer paramsPointer = null;
private Pointer greedyParamsPointer = null;
private Pointer beamParamsPointer = null;
public File modelDir() {
String modelDirPath = System.getenv("XDG_CACHE_HOME");
if (modelDirPath == null) {
modelDirPath = System.getProperty("user.home") + "/.cache";
}
return new File(modelDirPath, "whisper");
}
/**
* @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
*/
public void initContext(String modelPath) throws FileNotFoundException {
initContextImpl(modelPath, getContextDefaultParams());
}
/**
* @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
* @param params - params to use when initialising the context
*/
public void initContext(String modelPath, WhisperContextParams params) throws FileNotFoundException {
initContextImpl(modelPath, params);
}
private void initContextImpl(String modelPath, WhisperContextParams params) throws FileNotFoundException {
if (ctx != null) {
lib.whisper_free(ctx);
}
if (!modelPath.contains("/") && !modelPath.contains("\\")) {
if (!modelPath.endsWith(".bin")) {
modelPath = "ggml-" + modelPath.replace("-", ".") + ".bin";
}
modelPath = new File(modelDir(), modelPath).getAbsolutePath();
}
ctx = lib.whisper_init_from_file_with_params(modelPath, params);
if (ctx == null) {
throw new FileNotFoundException(modelPath);
}
}
/**
* Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
* Because this function allocates memory for the params, the caller must call either:
* - call `whisper_free_context_params()`
* - `Native.free(Pointer.nativeValue(pointer));`
*/
public WhisperContextParams getContextDefaultParams() {
paramsPointer = lib.whisper_context_default_params_by_ref();
WhisperContextParams params = new WhisperContextParams(paramsPointer);
params.read();
return params;
}
/**
* Provides default params which can be used with `whisper_full()` etc.
* Because this function allocates memory for the params, the caller must call either:
* - call `whisper_free_params()`
* - `Native.free(Pointer.nativeValue(pointer));`
*
* @param strategy - GREEDY
*/
public WhisperFullParams getFullDefaultParams(WhisperSamplingStrategy strategy) {
Pointer pointer;
// whisper_full_default_params_by_ref allocates memory which we need to delete, so only create max 1 pointer for each strategy.
if (strategy == WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY) {
if (greedyParamsPointer == null) {
greedyParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
}
pointer = greedyParamsPointer;
} else {
if (beamParamsPointer == null) {
beamParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
}
pointer = beamParamsPointer;
}
WhisperFullParams params = new WhisperFullParams(pointer);
params.read();
return params;
}
@Override
public void close() {
freeContext();
freeParams();
System.out.println("Whisper closed");
}
private void freeContext() {
if (ctx != null) {
lib.whisper_free(ctx);
}
}
private void freeParams() {
if (paramsPointer != null) {
Native.free(Pointer.nativeValue(paramsPointer));
paramsPointer = null;
}
if (greedyParamsPointer != null) {
Native.free(Pointer.nativeValue(greedyParamsPointer));
greedyParamsPointer = null;
}
if (beamParamsPointer != null) {
Native.free(Pointer.nativeValue(beamParamsPointer));
beamParamsPointer = null;
}
}
/**
* Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text.
* Not thread safe for same context
* Uses the specified decoding strategy to obtain the text.
*/
public String fullTranscribe(WhisperFullParams whisperParams, float[] audioData) throws IOException {
if (ctx == null) {
throw new IllegalStateException("Model not initialised");
}
if (lib.whisper_full(ctx, whisperParams, audioData, audioData.length) != 0) {
throw new IOException("Failed to process audio");
}
int nSegments = lib.whisper_full_n_segments(ctx);
StringBuilder str = new StringBuilder();
for (int i = 0; i < nSegments; i++) {
String text = lib.whisper_full_get_segment_text(ctx, i);
System.out.println("Segment:" + text);
str.append(text);
}
return str.toString().trim();
}
public List<WhisperSegment> fullTranscribeWithTime(WhisperFullParams whisperParams, float[] audioData) throws IOException {
if (ctx == null) {
throw new IllegalStateException("Model not initialised");
}
if (lib.whisper_full(ctx, whisperParams, audioData, audioData.length) != 0) {
throw new IOException("Failed to process audio");
}
int nSegments = lib.whisper_full_n_segments(ctx);
List<WhisperSegment> segments= new ArrayList<>(nSegments);
for (int i = 0; i < nSegments; i++) {
long t0 = lib.whisper_full_get_segment_t0(ctx, i);
String text = lib.whisper_full_get_segment_text(ctx, i);
long t1 = lib.whisper_full_get_segment_t1(ctx, i);
segments.add(new WhisperSegment(t0,t1,text));
}
return segments;
}
// public int getTextSegmentCount(Pointer ctx) {
// return lib.whisper_full_n_segments(ctx);
// }
// public String getTextSegment(Pointer ctx, int index) {
// return lib.whisper_full_get_segment_text(ctx, index);
// }
public String getSystemInfo() {
return lib.whisper_print_system_info();
}
public int benchMemcpy(int nthread) {
return lib.whisper_bench_memcpy(nthread);
}
public int benchGgmlMulMat(int nthread) {
return lib.whisper_bench_ggml_mul_mat(nthread);
}
}

396
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCppJnaLibrary.java Normal file
View File

@@ -0,0 +1,396 @@
package io.github.ggerganov.whispercpp;
import com.sun.jna.Library;
import com.sun.jna.Native;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.model.WhisperModelLoader;
import io.github.ggerganov.whispercpp.model.WhisperTokenData;
import io.github.ggerganov.whispercpp.params.WhisperContextParams;
import io.github.ggerganov.whispercpp.params.WhisperFullParams;
public interface WhisperCppJnaLibrary extends Library {
WhisperCppJnaLibrary instance = Native.load("whisper", WhisperCppJnaLibrary.class);
String whisper_print_system_info();
/**
* DEPRECATED. Allocate (almost) all memory needed for the model by loading from a file.
*
* @param path_model Path to the model file
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_from_file(String path_model);
/**
* Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
* Because this function allocates memory for the params, the caller must call either:
* - call `whisper_free_context_params()`
* - `Native.free(Pointer.nativeValue(pointer));`
*/
Pointer whisper_context_default_params_by_ref();
void whisper_free_context_params(Pointer params);
/**
* Allocate (almost) all memory needed for the model by loading from a file.
*
* @param path_model Path to the model file
* @param params Pointer to whisper_context_params
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_from_file_with_params(String path_model, WhisperContextParams params);
/**
* Allocate (almost) all memory needed for the model by loading from a buffer.
*
* @param buffer Model buffer
* @param buffer_size Size of the model buffer
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_from_buffer(Pointer buffer, int buffer_size);
/**
* Allocate (almost) all memory needed for the model using a model loader.
*
* @param loader Model loader
* @return Whisper context on success, null on failure
*/
Pointer whisper_init(WhisperModelLoader loader);
/**
* Allocate (almost) all memory needed for the model by loading from a file without allocating the state.
*
* @param path_model Path to the model file
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_from_file_no_state(String path_model);
/**
* Allocate (almost) all memory needed for the model by loading from a buffer without allocating the state.
*
* @param buffer Model buffer
* @param buffer_size Size of the model buffer
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_from_buffer_no_state(Pointer buffer, int buffer_size);
// Pointer whisper_init_from_buffer_no_state(Pointer buffer, long buffer_size);
/**
* Allocate (almost) all memory needed for the model using a model loader without allocating the state.
*
* @param loader Model loader
* @return Whisper context on success, null on failure
*/
Pointer whisper_init_no_state(WhisperModelLoader loader);
/**
* Allocate memory for the Whisper state.
*
* @param ctx Whisper context
* @return Whisper state on success, null on failure
*/
Pointer whisper_init_state(Pointer ctx);
/**
* Free all allocated memory associated with the Whisper context.
*
* @param ctx Whisper context
*/
void whisper_free(Pointer ctx);
/**
* Free all allocated memory associated with the Whisper state.
*
* @param state Whisper state
*/
void whisper_free_state(Pointer state);
/**
* Convert RAW PCM audio to log mel spectrogram.
* The resulting spectrogram is stored inside the default state of the provided whisper context.
*
* @param ctx - Pointer to a WhisperContext
* @return 0 on success
*/
int whisper_pcm_to_mel(Pointer ctx, final float[] samples, int n_samples, int n_threads);
/**
* @param ctx Pointer to a WhisperContext
* @param state Pointer to WhisperState
* @param n_samples
* @param n_threads
* @return 0 on success
*/
int whisper_pcm_to_mel_with_state(Pointer ctx, Pointer state, final float[] samples, int n_samples, int n_threads);
/**
* This can be used to set a custom log mel spectrogram inside the default state of the provided whisper context.
* Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
* n_mel must be 80
* @return 0 on success
*/
int whisper_set_mel(Pointer ctx, final float[] data, int n_len, int n_mel);
int whisper_set_mel_with_state(Pointer ctx, Pointer state, final float[] data, int n_len, int n_mel);
/**
* Run the Whisper encoder on the log mel spectrogram stored inside the default state in the provided whisper context.
* Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
* Offset can be used to specify the offset of the first frame in the spectrogram.
* @return 0 on success
*/
int whisper_encode(Pointer ctx, int offset, int n_threads);
int whisper_encode_with_state(Pointer ctx, Pointer state, int offset, int n_threads);
/**
* Run the Whisper decoder to obtain the logits and probabilities for the next token.
* Make sure to call whisper_encode() first.
* tokens + n_tokens is the provided context for the decoder.
* n_past is the number of tokens to use from previous decoder calls.
* Returns 0 on success
* TODO: add support for multiple decoders
*/
int whisper_decode(Pointer ctx, Pointer tokens, int n_tokens, int n_past, int n_threads);
/**
* @param ctx
* @param state
* @param tokens Pointer to int tokens
* @param n_tokens
* @param n_past
* @param n_threads
* @return
*/
int whisper_decode_with_state(Pointer ctx, Pointer state, Pointer tokens, int n_tokens, int n_past, int n_threads);
/**
* Convert the provided text into tokens.
* The tokens pointer must be large enough to hold the resulting tokens.
* Returns the number of tokens on success, no more than n_max_tokens
* Returns -1 on failure
* TODO: not sure if correct
*/
int whisper_tokenize(Pointer ctx, String text, Pointer tokens, int n_max_tokens);
/** Largest language id (i.e. number of available languages - 1) */
int whisper_lang_max_id();
/**
* @return the id of the specified language, returns -1 if not found.
* Examples:
* "de" -> 2
* "german" -> 2
*/
int whisper_lang_id(String lang);
/** @return the short string of the specified language id (e.g. 2 -> "de"), returns nullptr if not found */
String whisper_lang_str(int id);
/**
* Use mel data at offset_ms to try and auto-detect the spoken language.
* Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first
* Returns the top language id or negative on failure
* If not null, fills the lang_probs array with the probabilities of all languages
* The array must be whisper_lang_max_id() + 1 in size
*
* ref: https://github.com/openai/whisper/blob/main/whisper/decoding.py#L18-L69
*/
int whisper_lang_auto_detect(Pointer ctx, int offset_ms, int n_threads, float[] lang_probs);
int whisper_lang_auto_detect_with_state(Pointer ctx, Pointer state, int offset_ms, int n_threads, float[] lang_probs);
int whisper_n_len (Pointer ctx); // mel length
int whisper_n_len_from_state(Pointer state); // mel length
int whisper_n_vocab (Pointer ctx);
int whisper_n_text_ctx (Pointer ctx);
int whisper_n_audio_ctx (Pointer ctx);
int whisper_is_multilingual (Pointer ctx);
int whisper_model_n_vocab (Pointer ctx);
int whisper_model_n_audio_ctx (Pointer ctx);
int whisper_model_n_audio_state(Pointer ctx);
int whisper_model_n_audio_head (Pointer ctx);
int whisper_model_n_audio_layer(Pointer ctx);
int whisper_model_n_text_ctx (Pointer ctx);
int whisper_model_n_text_state (Pointer ctx);
int whisper_model_n_text_head (Pointer ctx);
int whisper_model_n_text_layer (Pointer ctx);
int whisper_model_n_mels (Pointer ctx);
int whisper_model_ftype (Pointer ctx);
int whisper_model_type (Pointer ctx);
/**
* Token logits obtained from the last call to whisper_decode().
* The logits for the last token are stored in the last row
* Rows: n_tokens
* Cols: n_vocab
*/
float[] whisper_get_logits (Pointer ctx);
float[] whisper_get_logits_from_state(Pointer state);
// Token Id -> String. Uses the vocabulary in the provided context
String whisper_token_to_str(Pointer ctx, int token);
String whisper_model_type_readable(Pointer ctx);
// Special tokens
int whisper_token_eot (Pointer ctx);
int whisper_token_sot (Pointer ctx);
int whisper_token_prev(Pointer ctx);
int whisper_token_solm(Pointer ctx);
int whisper_token_not (Pointer ctx);
int whisper_token_beg (Pointer ctx);
int whisper_token_lang(Pointer ctx, int lang_id);
// Task tokens
int whisper_token_translate (Pointer ctx);
int whisper_token_transcribe(Pointer ctx);
// Performance information from the default state.
void whisper_print_timings(Pointer ctx);
void whisper_reset_timings(Pointer ctx);
// Note: Even if `whisper_full_params is stripped back to just 4 ints, JNA throws "Invalid memory access"
// when `whisper_full_default_params()` tries to return a struct.
// WhisperFullParams whisper_full_default_params(int strategy);
/**
* Provides default params which can be used with `whisper_full()` etc.
* Because this function allocates memory for the params, the caller must call either:
* - call `whisper_free_params()`
* - `Native.free(Pointer.nativeValue(pointer));`
*
* @param strategy - WhisperSamplingStrategy.value
*/
Pointer whisper_full_default_params_by_ref(int strategy);
void whisper_free_params(Pointer params);
/**
* Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
* Not thread safe for same context
* Uses the specified decoding strategy to obtain the text.
*/
int whisper_full(Pointer ctx, WhisperFullParams params, final float[] samples, int n_samples);
int whisper_full_with_state(Pointer ctx, Pointer state, WhisperFullParams params, final float[] samples, int n_samples);
// Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
// Result is stored in the default state of the context
// Not thread safe if executed in parallel on the same context.
// It seems this approach can offer some speedup in some cases.
// However, the transcription accuracy can be worse at the beginning and end of each chunk.
int whisper_full_parallel(Pointer ctx, WhisperFullParams params, final float[] samples, int n_samples, int n_processors);
/**
* Number of generated text segments.
* A segment can be a few words, a sentence, or even a paragraph.
* @param ctx Pointer to WhisperContext
*/
int whisper_full_n_segments (Pointer ctx);
/**
* @param state Pointer to WhisperState
*/
int whisper_full_n_segments_from_state(Pointer state);
/**
* Language id associated with the context's default state.
* @param ctx Pointer to WhisperContext
*/
int whisper_full_lang_id(Pointer ctx);
/** Language id associated with the provided state */
int whisper_full_lang_id_from_state(Pointer state);
/**
* Convert RAW PCM audio to log mel spectrogram but applies a Phase Vocoder to speed up the audio x2.
* The resulting spectrogram is stored inside the default state of the provided whisper context.
* @return 0 on success
*/
int whisper_pcm_to_mel_phase_vocoder(Pointer ctx, final float[] samples, int n_samples, int n_threads);
int whisper_pcm_to_mel_phase_vocoder_with_state(Pointer ctx, Pointer state, final float[] samples, int n_samples, int n_threads);
/** Get the start time of the specified segment. */
long whisper_full_get_segment_t0(Pointer ctx, int i_segment);
/** Get the start time of the specified segment from the state. */
long whisper_full_get_segment_t0_from_state(Pointer state, int i_segment);
/** Get the end time of the specified segment. */
long whisper_full_get_segment_t1(Pointer ctx, int i_segment);
/** Get the end time of the specified segment from the state. */
long whisper_full_get_segment_t1_from_state(Pointer state, int i_segment);
/** Get the text of the specified segment. */
String whisper_full_get_segment_text(Pointer ctx, int i_segment);
/** Get the text of the specified segment from the state. */
String whisper_full_get_segment_text_from_state(Pointer state, int i_segment);
/** Get the number of tokens in the specified segment. */
int whisper_full_n_tokens(Pointer ctx, int i_segment);
/** Get the number of tokens in the specified segment from the state. */
int whisper_full_n_tokens_from_state(Pointer state, int i_segment);
/** Get the token text of the specified token in the specified segment. */
String whisper_full_get_token_text(Pointer ctx, int i_segment, int i_token);
/** Get the token text of the specified token in the specified segment from the state. */
String whisper_full_get_token_text_from_state(Pointer ctx, Pointer state, int i_segment, int i_token);
/** Get the token ID of the specified token in the specified segment. */
int whisper_full_get_token_id(Pointer ctx, int i_segment, int i_token);
/** Get the token ID of the specified token in the specified segment from the state. */
int whisper_full_get_token_id_from_state(Pointer state, int i_segment, int i_token);
/** Get token data for the specified token in the specified segment. */
WhisperTokenData whisper_full_get_token_data(Pointer ctx, int i_segment, int i_token);
/** Get token data for the specified token in the specified segment from the state. */
WhisperTokenData whisper_full_get_token_data_from_state(Pointer state, int i_segment, int i_token);
/** Get the probability of the specified token in the specified segment. */
float whisper_full_get_token_p(Pointer ctx, int i_segment, int i_token);
/** Get the probability of the specified token in the specified segment from the state. */
float whisper_full_get_token_p_from_state(Pointer state, int i_segment, int i_token);
/**
* Benchmark function for memcpy.
*
* @param nThreads Number of threads to use for the benchmark.
* @return The result of the benchmark.
*/
int whisper_bench_memcpy(int nThreads);
/**
* Benchmark function for memcpy as a string.
*
* @param nThreads Number of threads to use for the benchmark.
* @return The result of the benchmark as a string.
*/
String whisper_bench_memcpy_str(int nThreads);
/**
* Benchmark function for ggml_mul_mat.
*
* @param nThreads Number of threads to use for the benchmark.
* @return The result of the benchmark.
*/
int whisper_bench_ggml_mul_mat(int nThreads);
/**
* Benchmark function for ggml_mul_mat as a string.
*
* @param nThreads Number of threads to use for the benchmark.
* @return The result of the benchmark as a string.
*/
String whisper_bench_ggml_mul_mat_str(int nThreads);
}

47
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/bean/WhisperSegment.java Normal file
View File

@@ -0,0 +1,47 @@
package io.github.ggerganov.whispercpp.bean;
/**
* Created by litonglinux@qq.com on 10/21/2023_7:48 AM
*/
public class WhisperSegment {
private long start, end;
private String sentence;
public WhisperSegment() {
}
public WhisperSegment(long start, long end, String sentence) {
this.start = start;
this.end = end;
this.sentence = sentence;
}
public long getStart() {
return start;
}
public long getEnd() {
return end;
}
public String getSentence() {
return sentence;
}
public void setStart(long start) {
this.start = start;
}
public void setEnd(long end) {
this.end = end;
}
public void setSentence(String sentence) {
this.sentence = sentence;
}
@Override
public String toString() {
return "[" + start + " --> " + end + "]:" + sentence;
}
}

24
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/WhisperEncoderBeginCallback.java Normal file
View File

@@ -0,0 +1,24 @@
package io.github.ggerganov.whispercpp.callbacks;
import com.sun.jna.Callback;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.WhisperContext;
import io.github.ggerganov.whispercpp.model.WhisperState;
/**
* Callback before the encoder starts.
* If not null, called before the encoder starts.
* If it returns false, the computation is aborted.
*/
public interface WhisperEncoderBeginCallback extends Callback {
/**
* Callback method before the encoder starts.
*
* @param ctx The whisper context.
* @param state The whisper state.
* @param user_data User data.
* @return True if the computation should proceed, false otherwise.
*/
boolean callback(Pointer ctx, Pointer state, Pointer user_data);
}

25
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/WhisperLogitsFilterCallback.java Normal file
View File

@@ -0,0 +1,25 @@
package io.github.ggerganov.whispercpp.callbacks;
import com.sun.jna.Callback;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.model.WhisperTokenData;
/**
* Callback to filter logits.
* Can be used to modify the logits before sampling.
* If not null, called after applying temperature to logits.
*/
public interface WhisperLogitsFilterCallback extends Callback {
/**
* Callback method to filter logits.
*
* @param ctx The whisper context.
* @param state The whisper state.
* @param tokens The array of whisper_token_data.
* @param n_tokens The number of tokens.
* @param logits The array of logits.
* @param user_data User data.
*/
void callback(Pointer ctx, Pointer state, WhisperTokenData[] tokens, int n_tokens, float[] logits, Pointer user_data);
}

24
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/WhisperNewSegmentCallback.java Normal file
View File

@@ -0,0 +1,24 @@
package io.github.ggerganov.whispercpp.callbacks;
import com.sun.jna.Callback;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.WhisperContext;
import io.github.ggerganov.whispercpp.model.WhisperState;
/**
* Callback for the text segment.
* Called on every newly generated text segment.
* Use the whisper_full_...() functions to obtain the text segments.
*/
public interface WhisperNewSegmentCallback extends Callback {
/**
* Callback method for the text segment.
*
* @param ctx The whisper context.
* @param state The whisper state.
* @param n_new The number of newly generated text segments.
* @param user_data User data.
*/
void callback(Pointer ctx, Pointer state, int n_new, Pointer user_data);
}

22
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/WhisperProgressCallback.java Normal file
View File

@@ -0,0 +1,22 @@
package io.github.ggerganov.whispercpp.callbacks;
import com.sun.jna.Callback;
import com.sun.jna.Pointer;
import io.github.ggerganov.whispercpp.WhisperContext;
import io.github.ggerganov.whispercpp.model.WhisperState;
/**
* Callback for progress updates.
*/
public interface WhisperProgressCallback extends Callback {
/**
* Callback method for progress updates.
*
* @param ctx The whisper context.
* @param state The whisper state.
* @param progress The progress value.
* @param user_data User data.
*/
void callback(Pointer ctx, Pointer state, int progress, Pointer user_data);
}

4
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/ggml/GgmlTensor.java Normal file
View File

@@ -0,0 +1,4 @@
package io.github.ggerganov.whispercpp.ggml;
public class GgmlTensor {
}

18
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/ggml/GgmlType.java Normal file
View File

@@ -0,0 +1,18 @@
package io.github.ggerganov.whispercpp.ggml;
public enum GgmlType {
GGML_TYPE_F32,
GGML_TYPE_F16,
GGML_TYPE_Q4_0,
GGML_TYPE_Q4_1,
REMOVED_GGML_TYPE_Q4_2, // support has been removed
REMOVED_GGML_TYPE_Q4_3, // support has been removed
GGML_TYPE_Q5_0,
GGML_TYPE_Q5_1,
GGML_TYPE_Q8_0,
GGML_TYPE_Q8_1,
GGML_TYPE_I8,
GGML_TYPE_I16,
GGML_TYPE_I32,
GGML_TYPE_COUNT,
}

10
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/model/EModel.java Normal file
View File

@@ -0,0 +1,10 @@
package io.github.ggerganov.whispercpp.model;
public enum EModel {
MODEL_UNKNOWN,
MODEL_TINY,
MODEL_BASE,
MODEL_SMALL,
MODEL_MEDIUM,
MODEL_LARGE,
}

49
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/model/WhisperModel.java Normal file
View File

@@ -0,0 +1,49 @@
package io.github.ggerganov.whispercpp;
import io.github.ggerganov.whispercpp.ggml.GgmlTensor;
import io.github.ggerganov.whispercpp.model.EModel;
public class WhisperModel {
// EModel type = EModel.MODEL_UNKNOWN;
//
// WhisperHParams hparams;
// WhisperFilters filters;
//
// // encoder.positional_embedding
// GgmlTensor e_pe;
//
// // encoder.conv1
// GgmlTensor e_conv_1_w;
// GgmlTensor e_conv_1_b;
//
// // encoder.conv2
// GgmlTensor e_conv_2_w;
// GgmlTensor e_conv_2_b;
//
// // encoder.ln_post
// GgmlTensor e_ln_w;
// GgmlTensor e_ln_b;
//
// // decoder.positional_embedding
// GgmlTensor d_pe;
//
// // decoder.token_embedding
// GgmlTensor d_te;
//
// // decoder.ln
// GgmlTensor d_ln_w;
// GgmlTensor d_ln_b;
//
// std::vector<whisper_layer_encoder> layers_encoder;
// std::vector<whisper_layer_decoder> layers_decoder;
//
// // context
// struct ggml_context * ctx;
//
// // the model memory buffer is read-only and can be shared between processors
// std::vector<uint8_t> * buf;
//
// // tensors
// int n_loaded;
// Map<String, GgmlTensor> tensors;
}

62
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/model/WhisperModelLoader.java Normal file
View File

@@ -0,0 +1,62 @@
package io.github.ggerganov.whispercpp.model;
import com.sun.jna.Callback;
import com.sun.jna.Pointer;
import com.sun.jna.Structure;
public class WhisperModelLoader extends Structure {
public Pointer context;
public ReadFunction read;
public EOFFunction eof;
public CloseFunction close;
public static class ReadFunction implements Callback {
public Pointer invoke(Pointer ctx, Pointer output, int readSize) {
// TODO
return ctx;
}
}
public static class EOFFunction implements Callback {
public boolean invoke(Pointer ctx) {
// TODO
return false;
}
}
public static class CloseFunction implements Callback {
public void invoke(Pointer ctx) {
// TODO
}
}
// public WhisperModelLoader(Pointer p) {
// super(p);
// read = new ReadFunction();
// eof = new EOFFunction();
// close = new CloseFunction();
// read.setCallback(this);
// eof.setCallback(this);
// close.setCallback(this);
// read.write();
// eof.write();
// close.write();
// }
public WhisperModelLoader() {
super();
}
public interface ReadCallback extends Callback {
Pointer invoke(Pointer ctx, Pointer output, int readSize);
}
public interface EOFCallback extends Callback {
boolean invoke(Pointer ctx);
}
public interface CloseCallback extends Callback {
void invoke(Pointer ctx);
}
}

4
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/model/WhisperState.java Normal file
View File

@@ -0,0 +1,4 @@
package io.github.ggerganov.whispercpp.model;
public class WhisperState {
}

50
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/model/WhisperTokenData.java Normal file
View File

@@ -0,0 +1,50 @@
package io.github.ggerganov.whispercpp.model;
import com.sun.jna.Structure;
import java.util.Arrays;
import java.util.List;
/**
* Structure representing token data.
*/
public class WhisperTokenData extends Structure {
/** Token ID. */
public int id;
/** Forced timestamp token ID. */
public int tid;
/** Probability of the token. */
public float p;
/** Log probability of the token. */
public float plog;
/** Probability of the timestamp token. */
public float pt;
/** Sum of probabilities of all timestamp tokens. */
public float ptsum;
/**
* Start time of the token (token-level timestamp data).
* Do not use if you haven't computed token-level timestamps.
*/
public long t0;
/**
* End time of the token (token-level timestamp data).
* Do not use if you haven't computed token-level timestamps.
*/
public long t1;
/** Voice length of the token. */
public float vlen;
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("id", "tid", "p", "plog", "pt", "ptsum", "t0", "t1", "vlen");
}
}

19
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/BeamSearchParams.java Normal file
View File

@@ -0,0 +1,19 @@
package io.github.ggerganov.whispercpp.params;
import com.sun.jna.Structure;
import java.util.Arrays;
import java.util.List;
public class BeamSearchParams extends Structure {
/** ref: <a href="https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/transcribe.py#L265">...</a> */
public int beam_size;
/** ref: <a href="https://arxiv.org/pdf/2204.05424.pdf">...</a> */
public float patience;
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("beam_size", "patience");
}
}

30
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/CBool.java Normal file
View File

@@ -0,0 +1,30 @@
package io.github.ggerganov.whispercpp.params;
import com.sun.jna.IntegerType;
import java.util.function.BooleanSupplier;
public class CBool extends IntegerType implements BooleanSupplier {
public static final int SIZE = 1;
public static final CBool FALSE = new CBool(0);
public static final CBool TRUE = new CBool(1);
public CBool() {
this(0);
}
public CBool(long value) {
super(SIZE, value, true);
}
@Override
public boolean getAsBoolean() {
return intValue() == 1;
}
@Override
public String toString() {
return intValue() == 1 ? "true" : "false";
}
}

16
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/GreedyParams.java Normal file
View File

@@ -0,0 +1,16 @@
package io.github.ggerganov.whispercpp.params;
import com.sun.jna.Structure;
import java.util.Collections;
import java.util.List;
public class GreedyParams extends Structure {
/** <a href="https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/transcribe.py#L264">...</a> */
public int best_of;
@Override
protected List<String> getFieldOrder() {
return Collections.singletonList("best_of");
}
}

31
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java Normal file
View File

@@ -0,0 +1,31 @@
package io.github.ggerganov.whispercpp.params;
import com.sun.jna.*;
import java.util.Arrays;
import java.util.List;
/**
* Parameters for the whisper_init_from_file_with_params() function.
* If you change the order or add new parameters, make sure to update the default values in whisper.cpp:
* whisper_context_default_params()
*/
public class WhisperContextParams extends Structure {
public WhisperContextParams(Pointer p) {
super(p);
}
/** Use GPU for inference Number (default = true) */
public CBool use_gpu;
/** Use GPU for inference Number (default = true) */
public void useGpu(boolean enable) {
use_gpu = enable ? CBool.TRUE : CBool.FALSE;
}
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("use_gpu");
}
}

10
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFilters.java Normal file
View File

@@ -0,0 +1,10 @@
package io.github.ggerganov.whispercpp.params;
import java.util.List;
public class WhisperFilters {
int n_mel;
int n_fft;
List<Float> data;
}

331
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFullParams.java Normal file
View File

@@ -0,0 +1,331 @@
package io.github.ggerganov.whispercpp.params;
import com.sun.jna.*;
import io.github.ggerganov.whispercpp.callbacks.WhisperEncoderBeginCallback;
import io.github.ggerganov.whispercpp.callbacks.WhisperLogitsFilterCallback;
import io.github.ggerganov.whispercpp.callbacks.WhisperNewSegmentCallback;
import io.github.ggerganov.whispercpp.callbacks.WhisperProgressCallback;
import java.util.Arrays;
import java.util.List;
/**
* Parameters for the whisper_full() function.
* If you change the order or add new parameters, make sure to update the default values in whisper.cpp:
* whisper_full_default_params()
*/
public class WhisperFullParams extends Structure {
public WhisperFullParams(Pointer p) {
super(p);
// super(p, ALIGN_MSVC);
// super(p, ALIGN_GNUC);
}
/** Sampling strategy for whisper_full() function. */
public int strategy;
/** Number of threads. (default = 4) */
public int n_threads;
/** Maximum tokens to use from past text as a prompt for the decoder. (default = 16384) */
public int n_max_text_ctx;
/** Start offset in milliseconds. (default = 0) */
public int offset_ms;
/** Audio duration to process in milliseconds. (default = 0) */
public int duration_ms;
/** Translate flag. (default = false) */
public CBool translate;
/** The compliment of translateMode() */
public void transcribeMode() {
translate = CBool.FALSE;
}
/** The compliment of transcribeMode() */
public void translateMode() {
translate = CBool.TRUE;
}
/** Flag to indicate whether to use past transcription (if any) as an initial prompt for the decoder. (default = true) */
public CBool no_context;
/** Flag to indicate whether to use past transcription (if any) as an initial prompt for the decoder. (default = true) */
public void enableContext(boolean enable) {
no_context = enable ? CBool.FALSE : CBool.TRUE;
}
/** Generate timestamps or not? */
public CBool no_timestamps;
/** Flag to force single segment output (useful for streaming). (default = false) */
public CBool single_segment;
/** Flag to force single segment output (useful for streaming). (default = false) */
public void singleSegment(boolean single) {
single_segment = single ? CBool.TRUE : CBool.FALSE;
}
/** Flag to print special tokens (e.g., &lt;SOT>, &lt;EOT>, &lt;BEG>, etc.). (default = false) */
public CBool print_special;
/** Flag to print special tokens (e.g., &lt;SOT>, &lt;EOT>, &lt;BEG>, etc.). (default = false) */
public void printSpecial(boolean enable) {
print_special = enable ? CBool.TRUE : CBool.FALSE;
}
/** Flag to print progress information. (default = true) */
public CBool print_progress;
/** Flag to print progress information. (default = true) */
public void printProgress(boolean enable) {
print_progress = enable ? CBool.TRUE : CBool.FALSE;
}
/** Flag to print results from within whisper.cpp (avoid it, use callback instead). (default = true) */
public CBool print_realtime;
/** Flag to print results from within whisper.cpp (avoid it, use callback instead). (default = true) */
public void printRealtime(boolean enable) {
print_realtime = enable ? CBool.TRUE : CBool.FALSE;
}
/** Flag to print timestamps for each text segment when printing realtime. (default = true) */
public CBool print_timestamps;
/** Flag to print timestamps for each text segment when printing realtime. (default = true) */
public void printTimestamps(boolean enable) {
print_timestamps = enable ? CBool.TRUE : CBool.FALSE;
}
/** [EXPERIMENTAL] Flag to enable token-level timestamps. (default = false) */
public CBool token_timestamps;
/** [EXPERIMENTAL] Flag to enable token-level timestamps. (default = false) */
public void tokenTimestamps(boolean enable) {
token_timestamps = enable ? CBool.TRUE : CBool.FALSE;
}
/** [EXPERIMENTAL] Timestamp token probability threshold (~0.01). (default = 0.01) */
public float thold_pt;
/** [EXPERIMENTAL] Timestamp token sum probability threshold (~0.01). */
public float thold_ptsum;
/** Maximum segment length in characters. (default = 0) */
public int max_len;
/** Flag to split on word rather than on token (when used with max_len). (default = false) */
public CBool split_on_word;
/** Flag to split on word rather than on token (when used with max_len). (default = false) */
public void splitOnWord(boolean enable) {
split_on_word = enable ? CBool.TRUE : CBool.FALSE;
}
/** Maximum tokens per segment (0, default = no limit) */
public int max_tokens;
/** Flag to speed up the audio by 2x using Phase Vocoder. (default = false) */
public CBool speed_up;
/** Flag to speed up the audio by 2x using Phase Vocoder. (default = false) */
public void speedUp(boolean enable) {
speed_up = enable ? CBool.TRUE : CBool.FALSE;
}
/** Overwrite the audio context size (0 = use default). */
public int audio_ctx;
/** Enable tinydiarize (default = false) */
public CBool tdrz_enable;
/** Enable tinydiarize (default = false) */
public void tdrzEnable(boolean enable) {
tdrz_enable = enable ? CBool.TRUE : CBool.FALSE;
}
/** Tokens to provide to the whisper decoder as an initial prompt.
* These are prepended to any existing text context from a previous call. */
public String initial_prompt;
/** Prompt tokens. (int*) */
public Pointer prompt_tokens;
public void setPromptTokens(int[] tokens) {
Memory mem = new Memory(tokens.length * 4L);
mem.write(0, tokens, 0, tokens.length);
prompt_tokens = mem;
}
/** Number of prompt tokens. */
public int prompt_n_tokens;
/** Language for auto-detection.
* For auto-detection, set to `null`, `""`, or "auto". */
public String language;
/** Flag to indicate whether to detect language automatically. */
public CBool detect_language;
/** Flag to indicate whether to detect language automatically. */
public void detectLanguage(boolean enable) {
detect_language = enable ? CBool.TRUE : CBool.FALSE;
}
// Common decoding parameters.
/** Flag to suppress blank tokens. */
public CBool suppress_blank;
public void suppressBlanks(boolean enable) {
suppress_blank = enable ? CBool.TRUE : CBool.FALSE;
}
/** Flag to suppress non-speech tokens. */
public CBool suppress_non_speech_tokens;
/** Flag to suppress non-speech tokens. */
public void suppressNonSpeechTokens(boolean enable) {
suppress_non_speech_tokens = enable ? CBool.TRUE : CBool.FALSE;
}
/** Initial decoding temperature. */
public float temperature;
/** Maximum initial timestamp. */
public float max_initial_ts;
/** Length penalty. */
public float length_penalty;
// Fallback parameters.
/** Temperature increment. */
public float temperature_inc;
/** Entropy threshold (similar to OpenAI's "compression_ratio_threshold"). */
public float entropy_thold;
/** Log probability threshold. */
public float logprob_thold;
/** No speech threshold. */
public float no_speech_thold;
/** Greedy decoding parameters. */
public GreedyParams greedy;
/**
* Beam search decoding parameters.
*/
public BeamSearchParams beam_search;
public void setBestOf(int bestOf) {
if (greedy == null) {
greedy = new GreedyParams();
}
greedy.best_of = bestOf;
}
public void setBeamSize(int beamSize) {
if (beam_search == null) {
beam_search = new BeamSearchParams();
}
beam_search.beam_size = beamSize;
}
public void setBeamSizeAndPatience(int beamSize, float patience) {
if (beam_search == null) {
beam_search = new BeamSearchParams();
}
beam_search.beam_size = beamSize;
beam_search.patience = patience;
}
/**
* Callback for every newly generated text segment.
* WhisperNewSegmentCallback
*/
public Pointer new_segment_callback;
/**
* User data for the new_segment_callback.
*/
public Pointer new_segment_callback_user_data;
/**
* Callback on each progress update.
* WhisperProgressCallback
*/
public Pointer progress_callback;
/**
* User data for the progress_callback.
*/
public Pointer progress_callback_user_data;
/**
* Callback each time before the encoder starts.
* WhisperEncoderBeginCallback
*/
public Pointer encoder_begin_callback;
/**
* User data for the encoder_begin_callback.
*/
public Pointer encoder_begin_callback_user_data;
/**
* Callback by each decoder to filter obtained logits.
* WhisperLogitsFilterCallback
*/
public Pointer logits_filter_callback;
/**
* User data for the logits_filter_callback.
*/
public Pointer logits_filter_callback_user_data;
public void setNewSegmentCallback(WhisperNewSegmentCallback callback) {
new_segment_callback = CallbackReference.getFunctionPointer(callback);
}
public void setProgressCallback(WhisperProgressCallback callback) {
progress_callback = CallbackReference.getFunctionPointer(callback);
}
public void setEncoderBeginCallbackeginCallbackCallback(WhisperEncoderBeginCallback callback) {
encoder_begin_callback = CallbackReference.getFunctionPointer(callback);
}
public void setLogitsFilterCallback(WhisperLogitsFilterCallback callback) {
logits_filter_callback = CallbackReference.getFunctionPointer(callback);
}
/** Grammar stuff */
public Pointer grammar_rules;
public long n_grammar_rules;
public long i_start_rule;
public float grammar_penalty;
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("strategy", "n_threads", "n_max_text_ctx", "offset_ms", "duration_ms", "translate",
"no_context", "single_segment", "no_timestamps",
"print_special", "print_progress", "print_realtime", "print_timestamps", "token_timestamps",
"thold_pt", "thold_ptsum", "max_len", "split_on_word", "max_tokens", "speed_up", "audio_ctx",
"tdrz_enable", "initial_prompt", "prompt_tokens", "prompt_n_tokens", "language", "detect_language",
"suppress_blank", "suppress_non_speech_tokens", "temperature", "max_initial_ts", "length_penalty",
"temperature_inc", "entropy_thold", "logprob_thold", "no_speech_thold", "greedy", "beam_search",
"new_segment_callback", "new_segment_callback_user_data",
"progress_callback", "progress_callback_user_data",
"encoder_begin_callback", "encoder_begin_callback_user_data",
"logits_filter_callback", "logits_filter_callback_user_data",
"grammar_rules", "n_grammar_rules", "i_start_rule", "grammar_penalty");
}
}

15
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperHParams.java Normal file
View File

@@ -0,0 +1,15 @@
package io.github.ggerganov.whispercpp.params;
public class WhisperHParams {
int n_vocab = 51864;
int n_audio_ctx = 1500;
int n_audio_state = 384;
int n_audio_head = 6;
int n_audio_layer = 4;
int n_text_ctx = 448;
int n_text_state = 384;
int n_text_head = 6;
int n_text_layer = 4;
int n_mels = 80;
int ftype = 1;
}

10
whisper.cpp-1.5.2/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperSamplingStrategy.java Normal file
View File

@@ -0,0 +1,10 @@
package io.github.ggerganov.whispercpp.params;
/** Available sampling strategies */
public enum WhisperSamplingStrategy {
/** similar to OpenAI's GreedyDecoder */
WHISPER_SAMPLING_GREEDY,
/** similar to OpenAI's BeamSearchDecoder */
WHISPER_SAMPLING_BEAM_SEARCH
}

144
whisper.cpp-1.5.2/bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperCppTest.java Normal file
View File

@@ -0,0 +1,144 @@
package io.github.ggerganov.whispercpp;
import static org.junit.jupiter.api.Assertions.*;
import io.github.ggerganov.whispercpp.bean.WhisperSegment;
import io.github.ggerganov.whispercpp.params.CBool;
import io.github.ggerganov.whispercpp.params.WhisperFullParams;
import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
import org.junit.jupiter.api.BeforeAll;
import org.junit.jupiter.api.Test;
import javax.sound.sampled.AudioInputStream;
import javax.sound.sampled.AudioSystem;
import java.io.File;
import java.io.FileNotFoundException;
import java.util.List;
class WhisperCppTest {
private static WhisperCpp whisper = new WhisperCpp();
private static boolean modelInitialised = false;
@BeforeAll
static void init() throws FileNotFoundException {
// By default, models are loaded from ~/.cache/whisper/ and are usually named "ggml-${name}.bin"
// or you can provide the absolute path to the model file.
//String modelName = "../../models/ggml-tiny.bin";
String modelName = "../../models/ggml-tiny.en.bin";
try {
whisper.initContext(modelName);
//whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
//whisper.getJavaDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
modelInitialised = true;
} catch (FileNotFoundException ex) {
System.out.println("Model " + modelName + " not found");
}
}
@Test
void testGetDefaultFullParams_BeamSearch() {
// When
WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
// Then
assertEquals(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH.ordinal(), params.strategy);
assertNotEquals(0, params.n_threads);
assertEquals(16384, params.n_max_text_ctx);
assertFalse(params.translate);
assertEquals(0.01f, params.thold_pt);
assertEquals(5, params.beam_search.beam_size);
assertEquals(-1.0f, params.beam_search.patience);
}
@Test
void testGetDefaultFullParams_Greedy() {
// When
WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
// Then
assertEquals(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY.ordinal(), params.strategy);
assertNotEquals(0, params.n_threads);
assertEquals(16384, params.n_max_text_ctx);
assertEquals(5, params.greedy.best_of);
}
@Test
void testFullTranscribe() throws Exception {
if (!modelInitialised) {
System.out.println("Model not initialised, skipping test");
return;
}
// Given
File file = new File(System.getProperty("user.dir"), "../../samples/jfk.wav");
AudioInputStream audioInputStream = AudioSystem.getAudioInputStream(file);
byte[] b = new byte[audioInputStream.available()];
float[] floats = new float[b.length / 2];
//WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
params.print_progress = CBool.FALSE;
//params.initial_prompt = "and so my fellow Americans um, like";
try {
audioInputStream.read(b);
for (int i = 0, j = 0; i < b.length; i += 2, j++) {
int intSample = (int) (b[i + 1]) << 8 | (int) (b[i]) & 0xFF;
floats[j] = intSample / 32767.0f;
}
// When
String result = whisper.fullTranscribe(params, floats);
// Then
System.err.println(result);
assertEquals("And so my fellow Americans ask not what your country can do for you " +
"ask what you can do for your country.",
result.replace(",", ""));
} finally {
audioInputStream.close();
}
}
@Test
void testFullTranscribeWithTime() throws Exception {
if (!modelInitialised) {
System.out.println("Model not initialised, skipping test");
return;
}
// Given
File file = new File(System.getProperty("user.dir"), "../../samples/jfk.wav");
AudioInputStream audioInputStream = AudioSystem.getAudioInputStream(file);
byte[] b = new byte[audioInputStream.available()];
float[] floats = new float[b.length / 2];
//WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
params.print_progress = CBool.FALSE;
//params.initial_prompt = "and so my fellow Americans um, like";
try {
audioInputStream.read(b);
for (int i = 0, j = 0; i < b.length; i += 2, j++) {
int intSample = (int) (b[i + 1]) << 8 | (int) (b[i]) & 0xFF;
floats[j] = intSample / 32767.0f;
}
List<WhisperSegment> segments = whisper.fullTranscribeWithTime(params, floats);
assertTrue(segments.size() > 0, "The size of segments should be greater than 0");
for (WhisperSegment segment : segments) {
System.out.println(segment);
}
} finally {
audioInputStream.close();
}
}
}

17
whisper.cpp-1.5.2/bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperJnaLibraryTest.java Normal file
View File

@@ -0,0 +1,17 @@
package io.github.ggerganov.whispercpp;
import static org.junit.jupiter.api.Assertions.*;
import org.junit.jupiter.api.Test;
class WhisperJnaLibraryTest {
@Test
void testWhisperPrint_system_info() {
String systemInfo = WhisperCppJnaLibrary.instance.whisper_print_system_info();
// eg: "AVX = 1 | AVX2 = 1 | AVX512 = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0
// | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | VSX = 0 | COREML = 0 | "
System.out.println("System info: " + systemInfo);
assertTrue(systemInfo.length() > 10);
}
}