init

whisper.cpp-1.5.2/bindings/go/pkg/whisper/consts.go

@@ -0,0 +1,28 @@
+package whisper
+
+import (
+	"errors"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// ERRORS
+
+var (
+	ErrUnableToLoadModel    = errors.New("unable to load model")
+	ErrInternalAppError     = errors.New("internal application error")
+	ErrProcessingFailed     = errors.New("processing failed")
+	ErrUnsupportedLanguage  = errors.New("unsupported language")
+	ErrModelNotMultilingual = errors.New("model is not multilingual")
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// CONSTANTS
+
+// SampleRate is the sample rate of the audio data.
+const SampleRate = whisper.SampleRate
+
+// SampleBits is the number of bytes per sample.
+const SampleBits = whisper.SampleBits

whisper.cpp-1.5.2/bindings/go/pkg/whisper/context.go

@@ -0,0 +1,316 @@
+package whisper
+
+import (
+	"fmt"
+	"io"
+	"runtime"
+	"strings"
+	"time"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type context struct {
+	n      int
+	model  *model
+	params whisper.Params
+}
+
+// Make sure context adheres to the interface
+var _ Context = (*context)(nil)
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+func newContext(model *model, params whisper.Params) (Context, error) {
+	context := new(context)
+	context.model = model
+	context.params = params
+
+	// Return success
+	return context, nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Set the language to use for speech recognition.
+func (context *context) SetLanguage(lang string) error {
+	if context.model.ctx == nil {
+		return ErrInternalAppError
+	}
+	if !context.model.IsMultilingual() {
+		return ErrModelNotMultilingual
+	}
+
+	if lang == "auto" {
+		context.params.SetLanguage(-1)
+	} else if id := context.model.ctx.Whisper_lang_id(lang); id < 0 {
+		return ErrUnsupportedLanguage
+	} else if err := context.params.SetLanguage(id); err != nil {
+		return err
+	}
+	// Return success
+	return nil
+}
+
+func (context *context) IsMultilingual() bool {
+	return context.model.IsMultilingual()
+}
+
+// Get language
+func (context *context) Language() string {
+	id := context.params.Language()
+	if id == -1 {
+		return "auto"
+	}
+	return whisper.Whisper_lang_str(context.params.Language())
+}
+
+// Set translate flag
+func (context *context) SetTranslate(v bool) {
+	context.params.SetTranslate(v)
+}
+
+// Set speedup flag
+func (context *context) SetSpeedup(v bool) {
+	context.params.SetSpeedup(v)
+}
+
+func (context *context) SetSplitOnWord(v bool) {
+	context.params.SetSplitOnWord(v)
+}
+
+// Set number of threads to use
+func (context *context) SetThreads(v uint) {
+	context.params.SetThreads(int(v))
+}
+
+// Set time offset
+func (context *context) SetOffset(v time.Duration) {
+	context.params.SetOffset(int(v.Milliseconds()))
+}
+
+// Set duration of audio to process
+func (context *context) SetDuration(v time.Duration) {
+	context.params.SetDuration(int(v.Milliseconds()))
+}
+
+// Set timestamp token probability threshold (~0.01)
+func (context *context) SetTokenThreshold(t float32) {
+	context.params.SetTokenThreshold(t)
+}
+
+// Set timestamp token sum probability threshold (~0.01)
+func (context *context) SetTokenSumThreshold(t float32) {
+	context.params.SetTokenSumThreshold(t)
+}
+
+// Set max segment length in characters
+func (context *context) SetMaxSegmentLength(n uint) {
+	context.params.SetMaxSegmentLength(int(n))
+}
+
+// Set token timestamps flag
+func (context *context) SetTokenTimestamps(b bool) {
+	context.params.SetTokenTimestamps(b)
+}
+
+// Set max tokens per segment (0 = no limit)
+func (context *context) SetMaxTokensPerSegment(n uint) {
+	context.params.SetMaxTokensPerSegment(int(n))
+}
+
+// Set audio encoder context
+func (context *context) SetAudioCtx(n uint) {
+	context.params.SetAudioCtx(int(n))
+}
+
+// ResetTimings resets the mode timings. Should be called before processing
+func (context *context) ResetTimings() {
+	context.model.ctx.Whisper_reset_timings()
+}
+
+// PrintTimings prints the model timings to stdout.
+func (context *context) PrintTimings() {
+	context.model.ctx.Whisper_print_timings()
+}
+
+// SystemInfo returns the system information
+func (context *context) SystemInfo() string {
+	return fmt.Sprintf("system_info: n_threads = %d / %d | %s\n",
+		context.params.Threads(),
+		runtime.NumCPU(),
+		whisper.Whisper_print_system_info(),
+	)
+}
+
+// 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 probabilities of all languages.
+func (context *context) WhisperLangAutoDetect(offset_ms int, n_threads int) ([]float32, error) {
+	langProbs, err := context.model.ctx.Whisper_lang_auto_detect(offset_ms, n_threads)
+	if err != nil {
+		return nil, err
+	}
+	return langProbs, nil
+}
+
+// Process new sample data and return any errors
+func (context *context) Process(
+	data []float32,
+	callNewSegment SegmentCallback,
+	callProgress ProgressCallback,
+) error {
+	if context.model.ctx == nil {
+		return ErrInternalAppError
+	}
+	// If the callback is defined then we force on single_segment mode
+	if callNewSegment != nil {
+		context.params.SetSingleSegment(true)
+	}
+
+	// We don't do parallel processing at the moment
+	processors := 0
+	if processors > 1 {
+		if err := context.model.ctx.Whisper_full_parallel(context.params, data, processors, nil, func(new int) {
+			if callNewSegment != nil {
+				num_segments := context.model.ctx.Whisper_full_n_segments()
+				s0 := num_segments - new
+				for i := s0; i < num_segments; i++ {
+					callNewSegment(toSegment(context.model.ctx, i))
+				}
+			}
+		}); err != nil {
+			return err
+		}
+	} else if err := context.model.ctx.Whisper_full(context.params, data, nil, func(new int) {
+		if callNewSegment != nil {
+			num_segments := context.model.ctx.Whisper_full_n_segments()
+			s0 := num_segments - new
+			for i := s0; i < num_segments; i++ {
+				callNewSegment(toSegment(context.model.ctx, i))
+			}
+		}
+	}, func(progress int) {
+		if callProgress != nil {
+			callProgress(progress)
+		}
+	}); err != nil {
+		return err
+	}
+
+	// Return success
+	return nil
+}
+
+// Return the next segment of tokens
+func (context *context) NextSegment() (Segment, error) {
+	if context.model.ctx == nil {
+		return Segment{}, ErrInternalAppError
+	}
+	if context.n >= context.model.ctx.Whisper_full_n_segments() {
+		return Segment{}, io.EOF
+	}
+
+	// Populate result
+	result := toSegment(context.model.ctx, context.n)
+
+	// Increment the cursor
+	context.n++
+
+	// Return success
+	return result, nil
+}
+
+// Test for text tokens
+func (context *context) IsText(t Token) bool {
+	switch {
+	case context.IsBEG(t):
+		return false
+	case context.IsSOT(t):
+		return false
+	case whisper.Token(t.Id) >= context.model.ctx.Whisper_token_eot():
+		return false
+	case context.IsPREV(t):
+		return false
+	case context.IsSOLM(t):
+		return false
+	case context.IsNOT(t):
+		return false
+	default:
+		return true
+	}
+}
+
+// Test for "begin" token
+func (context *context) IsBEG(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_beg()
+}
+
+// Test for "start of transcription" token
+func (context *context) IsSOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_sot()
+}
+
+// Test for "end of transcription" token
+func (context *context) IsEOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_eot()
+}
+
+// Test for "start of prev" token
+func (context *context) IsPREV(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_prev()
+}
+
+// Test for "start of lm" token
+func (context *context) IsSOLM(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_solm()
+}
+
+// Test for "No timestamps" token
+func (context *context) IsNOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_not()
+}
+
+// Test for token associated with a specific language
+func (context *context) IsLANG(t Token, lang string) bool {
+	if id := context.model.ctx.Whisper_lang_id(lang); id >= 0 {
+		return whisper.Token(t.Id) == context.model.ctx.Whisper_token_lang(id)
+	} else {
+		return false
+	}
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PRIVATE METHODS
+
+func toSegment(ctx *whisper.Context, n int) Segment {
+	return Segment{
+		Num:    n,
+		Text:   strings.TrimSpace(ctx.Whisper_full_get_segment_text(n)),
+		Start:  time.Duration(ctx.Whisper_full_get_segment_t0(n)) * time.Millisecond * 10,
+		End:    time.Duration(ctx.Whisper_full_get_segment_t1(n)) * time.Millisecond * 10,
+		Tokens: toTokens(ctx, n),
+	}
+}
+
+func toTokens(ctx *whisper.Context, n int) []Token {
+	result := make([]Token, ctx.Whisper_full_n_tokens(n))
+	for i := 0; i < len(result); i++ {
+		data := ctx.Whisper_full_get_token_data(n, i)
+
+		result[i] = Token{
+			Id:    int(ctx.Whisper_full_get_token_id(n, i)),
+			Text:  ctx.Whisper_full_get_token_text(n, i),
+			P:     ctx.Whisper_full_get_token_p(n, i),
+			Start: time.Duration(data.T0()) * time.Millisecond * 10,
+			End:   time.Duration(data.T1()) * time.Millisecond * 10,
+		}
+	}
+	return result
+}

whisper.cpp-1.5.2/bindings/go/pkg/whisper/context_test.go

@@ -0,0 +1,55 @@
+package whisper_test
+
+import (
+	"os"
+	"testing"
+
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	assert "github.com/stretchr/testify/assert"
+)
+
+const (
+	ModelPath  = "../../models/ggml-tiny.bin"
+	SamplePath = "../../samples/jfk.wav"
+)
+
+func Test_Whisper_000(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
+	model, err := whisper.New(ModelPath)
+	assert.NoError(err)
+	assert.NotNil(model)
+	assert.NoError(model.Close())
+
+	t.Log("languages=", model.Languages())
+}
+
+func Test_Whisper_001(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
+	model, err := whisper.New(ModelPath)
+	assert.NoError(err)
+	assert.NotNil(model)
+	defer model.Close()
+
+	// Get context for decoding
+	ctx, err := model.NewContext()
+	assert.NoError(err)
+	assert.NotNil(ctx)
+
+}

whisper.cpp-1.5.2/bindings/go/pkg/whisper/doc.go

@@ -0,0 +1,4 @@
+/*
+This is the higher-level speech-to-text whisper.cpp API for go
+*/
+package whisper

whisper.cpp-1.5.2/bindings/go/pkg/whisper/interface.go

@@ -0,0 +1,99 @@
+package whisper
+
+import (
+	"io"
+	"time"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+// SegmentCallback is the callback function for processing segments in real
+// time. It is called during the Process function
+type SegmentCallback func(Segment)
+
+// ProgressCallback is the callback function for reporting progress during
+// processing. It is called during the Process function
+type ProgressCallback func(int)
+
+// Model is the interface to a whisper model. Create a new model with the
+// function whisper.New(string)
+type Model interface {
+	io.Closer
+
+	// Return a new speech-to-text context.
+	NewContext() (Context, error)
+
+	// Return true if the model is multilingual.
+	IsMultilingual() bool
+
+	// Return all languages supported.
+	Languages() []string
+}
+
+// Context is the speach recognition context.
+type Context interface {
+	SetLanguage(string) error // Set the language to use for speech recognition, use "auto" for auto detect language.
+	SetTranslate(bool)        // Set translate flag
+	IsMultilingual() bool     // Return true if the model is multilingual.
+	Language() string         // Get language
+
+	SetOffset(time.Duration)      // Set offset
+	SetDuration(time.Duration)    // Set duration
+	SetThreads(uint)              // Set number of threads to use
+	SetSpeedup(bool)              // Set speedup flag
+	SetSplitOnWord(bool)          // Set split on word flag
+	SetTokenThreshold(float32)    // Set timestamp token probability threshold
+	SetTokenSumThreshold(float32) // Set timestamp token sum probability threshold
+	SetMaxSegmentLength(uint)     // Set max segment length in characters
+	SetTokenTimestamps(bool)      // Set token timestamps flag
+	SetMaxTokensPerSegment(uint)  // Set max tokens per segment (0 = no limit)
+	SetAudioCtx(uint)             // Set audio encoder context
+
+	// Process mono audio data and return any errors.
+	// If defined, newly generated segments are passed to the
+	// callback function during processing.
+	Process([]float32, SegmentCallback, ProgressCallback) error
+
+	// After process is called, return segments until the end of the stream
+	// is reached, when io.EOF is returned.
+	NextSegment() (Segment, error)
+
+	IsBEG(Token) bool          // Test for "begin" token
+	IsSOT(Token) bool          // Test for "start of transcription" token
+	IsEOT(Token) bool          // Test for "end of transcription" token
+	IsPREV(Token) bool         // Test for "start of prev" token
+	IsSOLM(Token) bool         // Test for "start of lm" token
+	IsNOT(Token) bool          // Test for "No timestamps" token
+	IsLANG(Token, string) bool // Test for token associated with a specific language
+	IsText(Token) bool         // Test for text token
+
+	// Timings
+	PrintTimings()
+	ResetTimings()
+
+	SystemInfo() string
+}
+
+// Segment is the text result of a speech recognition.
+type Segment struct {
+	// Segment Number
+	Num int
+
+	// Time beginning and end timestamps for the segment.
+	Start, End time.Duration
+
+	// The text of the segment.
+	Text string
+
+	// The tokens of the segment.
+	Tokens []Token
+}
+
+// Token is a text or special token
+type Token struct {
+	Id         int
+	Text       string
+	P          float32
+	Start, End time.Duration
+}

whisper.cpp-1.5.2/bindings/go/pkg/whisper/model.go

@@ -0,0 +1,101 @@
+package whisper
+
+import (
+	"fmt"
+	"os"
+	"runtime"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type model struct {
+	path string
+	ctx  *whisper.Context
+}
+
+// Make sure model adheres to the interface
+var _ Model = (*model)(nil)
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+func New(path string) (Model, error) {
+	model := new(model)
+	if _, err := os.Stat(path); err != nil {
+		return nil, err
+	} else if ctx := whisper.Whisper_init(path); ctx == nil {
+		return nil, ErrUnableToLoadModel
+	} else {
+		model.ctx = ctx
+		model.path = path
+	}
+
+	// Return success
+	return model, nil
+}
+
+func (model *model) Close() error {
+	if model.ctx != nil {
+		model.ctx.Whisper_free()
+	}
+
+	// Release resources
+	model.ctx = nil
+
+	// Return success
+	return nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// STRINGIFY
+
+func (model *model) String() string {
+	str := "<whisper.model"
+	if model.ctx != nil {
+		str += fmt.Sprintf(" model=%q", model.path)
+	}
+	return str + ">"
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Return true if model is multilingual (language and translation options are supported)
+func (model *model) IsMultilingual() bool {
+	return model.ctx.Whisper_is_multilingual() != 0
+}
+
+// Return all recognized languages. Initially it is set to auto-detect
+func (model *model) Languages() []string {
+	result := make([]string, 0, whisper.Whisper_lang_max_id())
+	for i := 0; i < whisper.Whisper_lang_max_id(); i++ {
+		str := whisper.Whisper_lang_str(i)
+		if model.ctx.Whisper_lang_id(str) >= 0 {
+			result = append(result, str)
+		}
+	}
+	return result
+}
+
+func (model *model) NewContext() (Context, error) {
+	if model.ctx == nil {
+		return nil, ErrInternalAppError
+	}
+
+	// Create new context
+	params := model.ctx.Whisper_full_default_params(whisper.SAMPLING_GREEDY)
+	params.SetTranslate(false)
+	params.SetPrintSpecial(false)
+	params.SetPrintProgress(false)
+	params.SetPrintRealtime(false)
+	params.SetPrintTimestamps(false)
+	params.SetThreads(runtime.NumCPU())
+	params.SetNoContext(true)
+
+	// Return new context
+	return newContext(model, params)
+}