feat: elevenlabs sound-generation api (#3355)

* initial version of elevenlabs compatible soundgeneration api and cli command Signed-off-by: Dave Lee <dave@gray101.com> * minor cleanup Signed-off-by: Dave Lee <dave@gray101.com> * restore TTS, add test Signed-off-by: Dave Lee <dave@gray101.com> * remove stray s Signed-off-by: Dave Lee <dave@gray101.com> * fix Signed-off-by: Dave Lee <dave@gray101.com> --------- Signed-off-by: Dave Lee <dave@gray101.com> Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com> Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
2025-05-28 06:25:00 +00:00 · 2024-08-23 20:20:28 -04:00 · 2024-08-23 20:20:28 -04:00 · 81ae92f017
commit 81ae92f017
parent 84d6e5a987
20 changed files with 450 additions and 37 deletions
--- a/core/backend/llm.go
+++ b/core/backend/llm.go
@ -87,7 +87,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			case string:
 				protoMessages[i].Content = ct
 			default:
-				return nil, fmt.Errorf("Unsupported type for schema.Message.Content for inference: %T", ct)
+				return nil, fmt.Errorf("unsupported type for schema.Message.Content for inference: %T", ct)
 			}
 		}
 	}
--- a/core/backend/soundgeneration.go
+++ b/core/backend/soundgeneration.go
@ -0,0 +1,74 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+func SoundGeneration(
+	backend string,
+	modelFile string,
+	text string,
+	duration *float32,
+	temperature *float32,
+	doSample *bool,
+	sourceFile *string,
+	sourceDivisor *int32,
+	loader *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig,
+) (string, *proto.Result, error) {
+	if backend == "" {
+		return "", nil, fmt.Errorf("backend is a required parameter")
+	}
+
+	grpcOpts := gRPCModelOpts(backendConfig)
+	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
+		model.WithBackendString(backend),
+		model.WithModel(modelFile),
+		model.WithContext(appConfig.Context),
+		model.WithAssetDir(appConfig.AssetsDestination),
+		model.WithLoadGRPCLoadModelOpts(grpcOpts),
+	})
+
+	soundGenModel, err := loader.BackendLoader(opts...)
+	if err != nil {
+		return "", nil, err
+	}
+
+	if soundGenModel == nil {
+		return "", nil, fmt.Errorf("could not load sound generation model")
+	}
+
+	if err := os.MkdirAll(appConfig.AudioDir, 0750); err != nil {
+		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
+	}
+
+	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "sound_generation", ".wav")
+	filePath := filepath.Join(appConfig.AudioDir, fileName)
+
+	res, err := soundGenModel.SoundGeneration(context.Background(), &proto.SoundGenerationRequest{
+		Text:        text,
+		Model:       modelFile,
+		Dst:         filePath,
+		Sample:      doSample,
+		Duration:    duration,
+		Temperature: temperature,
+		Src:         sourceFile,
+		SrcDivisor:  sourceDivisor,
+	})
+
+	// return RPC error if any
+	if !res.Success {
+		return "", nil, fmt.Errorf(res.Message)
+	}
+
+	return filePath, res, err
+}
--- a/core/backend/tts.go
+++ b/core/backend/tts.go
@ -9,31 +9,15 @@ import (
 	"github.com/mudler/LocalAI/core/config"

 	"github.com/mudler/LocalAI/pkg/grpc/proto"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/utils"
 )

-func generateUniqueFileName(dir, baseName, ext string) string {
-	counter := 1
-	fileName := baseName + ext
-
-	for {
-		filePath := filepath.Join(dir, fileName)
-		_, err := os.Stat(filePath)
-		if os.IsNotExist(err) {
-			return fileName
-		}
-
-		counter++
-		fileName = fmt.Sprintf("%s_%d%s", baseName, counter, ext)
-	}
-}
-
 func ModelTTS(
 	backend,
 	text,
 	modelFile,
-	voice ,
+	voice,
 	language string,
 	loader *model.ModelLoader,
 	appConfig *config.ApplicationConfig,
@ -66,7 +50,7 @@ func ModelTTS(
 		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
 	}

-	fileName := generateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
+	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
 	filePath := filepath.Join(appConfig.AudioDir, fileName)

 	// If the model file is not empty, we pass it joined with the model path
@ -88,10 +72,10 @@ func ModelTTS(
 	}

 	res, err := ttsModel.TTS(context.Background(), &proto.TTSRequest{
-		Text:  text,
-		Model: modelPath,
-		Voice: voice,
-		Dst:   filePath,
+		Text:     text,
+		Model:    modelPath,
+		Voice:    voice,
+		Dst:      filePath,
 		Language: &language,
 	})

--- a/core/cli/cli.go
+++ b/core/cli/cli.go
@ -8,12 +8,13 @@ import (
 var CLI struct {
 	cliContext.Context `embed:""`

-	Run        RunCMD        `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
-	Federated  FederatedCLI  `cmd:"" help:"Run LocalAI in federated mode"`
-	Models     ModelsCMD     `cmd:"" help:"Manage LocalAI models and definitions"`
-	TTS        TTSCMD        `cmd:"" help:"Convert text to speech"`
-	Transcript TranscriptCMD `cmd:"" help:"Convert audio to text"`
-	Worker     worker.Worker `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
-	Util       UtilCMD       `cmd:"" help:"Utility commands"`
-	Explorer   ExplorerCMD   `cmd:"" help:"Run p2p explorer"`
+	Run             RunCMD             `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
+	Federated       FederatedCLI       `cmd:"" help:"Run LocalAI in federated mode"`
+	Models          ModelsCMD          `cmd:"" help:"Manage LocalAI models and definitions"`
+	TTS             TTSCMD             `cmd:"" help:"Convert text to speech"`
+	SoundGeneration SoundGenerationCMD `cmd:"" help:"Generates audio files from text or audio"`
+	Transcript      TranscriptCMD      `cmd:"" help:"Convert audio to text"`
+	Worker          worker.Worker      `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
+	Util            UtilCMD            `cmd:"" help:"Utility commands"`
+	Explorer        ExplorerCMD        `cmd:"" help:"Run p2p explorer"`
 }
--- a/core/cli/soundgeneration.go
+++ b/core/cli/soundgeneration.go
@ -0,0 +1,110 @@
+package cli
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	"github.com/mudler/LocalAI/core/backend"
+	cliContext "github.com/mudler/LocalAI/core/cli/context"
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/pkg/model"
+	"github.com/rs/zerolog/log"
+)
+
+type SoundGenerationCMD struct {
+	Text []string `arg:""`
+
+	Backend                string   `short:"b" required:"" help:"Backend to run the SoundGeneration model"`
+	Model                  string   `short:"m" required:"" help:"Model name to run the SoundGeneration"`
+	Duration               string   `short:"d" help:"If specified, the length of audio to generate in seconds"`
+	Temperature            string   `short:"t" help:"If specified, the temperature of the generation"`
+	InputFile              string   `short:"i" help:"If specified, the input file to condition generation upon"`
+	InputFileSampleDivisor string   `short:"f" help:"If InputFile and this divisor is specified, the first portion of the sample file will be used"`
+	DoSample               bool     `short:"s" default:"true" help:"Enables sampling from the model. Better quality at the cost of speed. Defaults to enabled."`
+	OutputFile             string   `short:"o" type:"path" help:"The path to write the output wav file"`
+	ModelsPath             string   `env:"LOCALAI_MODELS_PATH,MODELS_PATH" type:"path" default:"${basepath}/models" help:"Path containing models used for inferencing" group:"storage"`
+	BackendAssetsPath      string   `env:"LOCALAI_BACKEND_ASSETS_PATH,BACKEND_ASSETS_PATH" type:"path" default:"/tmp/localai/backend_data" help:"Path used to extract libraries that are required by some of the backends in runtime" group:"storage"`
+	ExternalGRPCBackends   []string `env:"LOCALAI_EXTERNAL_GRPC_BACKENDS,EXTERNAL_GRPC_BACKENDS" help:"A list of external grpc backends" group:"backends"`
+}
+
+func parseToFloat32Ptr(input string) *float32 {
+	f, err := strconv.ParseFloat(input, 32)
+	if err != nil {
+		return nil
+	}
+	f2 := float32(f)
+	return &f2
+}
+
+func parseToInt32Ptr(input string) *int32 {
+	i, err := strconv.ParseInt(input, 10, 32)
+	if err != nil {
+		return nil
+	}
+	i2 := int32(i)
+	return &i2
+}
+
+func (t *SoundGenerationCMD) Run(ctx *cliContext.Context) error {
+	outputFile := t.OutputFile
+	outputDir := t.BackendAssetsPath
+	if outputFile != "" {
+		outputDir = filepath.Dir(outputFile)
+	}
+
+	text := strings.Join(t.Text, " ")
+
+	externalBackends := make(map[string]string)
+	// split ":" to get backend name and the uri
+	for _, v := range t.ExternalGRPCBackends {
+		backend := v[:strings.IndexByte(v, ':')]
+		uri := v[strings.IndexByte(v, ':')+1:]
+		externalBackends[backend] = uri
+		fmt.Printf("TMP externalBackends[%q]=%q\n\n", backend, uri)
+	}
+
+	opts := &config.ApplicationConfig{
+		ModelPath:            t.ModelsPath,
+		Context:              context.Background(),
+		AudioDir:             outputDir,
+		AssetsDestination:    t.BackendAssetsPath,
+		ExternalGRPCBackends: externalBackends,
+	}
+	ml := model.NewModelLoader(opts.ModelPath)
+
+	defer func() {
+		err := ml.StopAllGRPC()
+		if err != nil {
+			log.Error().Err(err).Msg("unable to stop all grpc processes")
+		}
+	}()
+
+	options := config.BackendConfig{}
+	options.SetDefaults()
+
+	var inputFile *string
+	if t.InputFile != "" {
+		inputFile = &t.InputFile
+	}
+
+	filePath, _, err := backend.SoundGeneration(t.Backend, t.Model, text,
+		parseToFloat32Ptr(t.Duration), parseToFloat32Ptr(t.Temperature), &t.DoSample,
+		inputFile, parseToInt32Ptr(t.InputFileSampleDivisor), ml, opts, options)
+
+	if err != nil {
+		return err
+	}
+	if outputFile != "" {
+		if err := os.Rename(filePath, outputFile); err != nil {
+			return err
+		}
+		fmt.Printf("Generate file %s\n", outputFile)
+	} else {
+		fmt.Printf("Generate file %s\n", filePath)
+	}
+	return nil
+}
--- a/core/http/endpoints/elevenlabs/soundgeneration.go
+++ b/core/http/endpoints/elevenlabs/soundgeneration.go
@ -0,0 +1,65 @@
+package elevenlabs
+
+import (
+	"github.com/gofiber/fiber/v2"
+	"github.com/mudler/LocalAI/core/backend"
+	"github.com/mudler/LocalAI/core/config"
+	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/model"
+	"github.com/rs/zerolog/log"
+)
+
+// SoundGenerationEndpoint is the ElevenLabs SoundGeneration endpoint https://elevenlabs.io/docs/api-reference/sound-generation
+// @Summary Generates audio from the input text.
+// @Param request body schema.ElevenLabsSoundGenerationRequest true "query params"
+// @Success 200 {string} binary	 "Response"
+// @Router /v1/sound-generation [post]
+func SoundGenerationEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		input := new(schema.ElevenLabsSoundGenerationRequest)
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+
+		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.ModelID, false)
+		if err != nil {
+			modelFile = input.ModelID
+			log.Warn().Str("ModelID", input.ModelID).Msg("Model not found in context")
+		}
+
+		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
+			config.LoadOptionDebug(appConfig.Debug),
+			config.LoadOptionThreads(appConfig.Threads),
+			config.LoadOptionContextSize(appConfig.ContextSize),
+			config.LoadOptionF16(appConfig.F16),
+		)
+		if err != nil {
+			modelFile = input.ModelID
+			log.Warn().Str("Request ModelID", input.ModelID).Err(err).Msg("error during LoadBackendConfigFileByName, using request ModelID")
+		} else {
+			if input.ModelID != "" {
+				modelFile = input.ModelID
+			} else {
+				modelFile = cfg.Model
+			}
+		}
+		log.Debug().Str("modelFile", "modelFile").Str("backend", cfg.Backend).Msg("Sound Generation Request about to be sent to backend")
+
+		if input.Duration != nil {
+			log.Debug().Float32("duration", *input.Duration).Msg("duration set")
+		}
+		if input.Temperature != nil {
+			log.Debug().Float32("temperature", *input.Temperature).Msg("temperature set")
+		}
+
+		// TODO: Support uploading files?
+		filePath, _, err := backend.SoundGeneration(cfg.Backend, modelFile, input.Text, input.Duration, input.Temperature, input.DoSample, nil, nil, ml, appConfig, *cfg)
+		if err != nil {
+			return err
+		}
+		return c.Download(filePath)
+
+	}
+}
--- a/core/http/routes/elevenlabs.go
+++ b/core/http/routes/elevenlabs.go
@ -16,4 +16,6 @@ func RegisterElevenLabsRoutes(app *fiber.App,
 	// Elevenlabs
 	app.Post("/v1/text-to-speech/:voice-id", auth, elevenlabs.TTSEndpoint(cl, ml, appConfig))

+	app.Post("/v1/sound-generation", auth, elevenlabs.SoundGenerationEndpoint(cl, ml, appConfig))
+
 }
--- a/core/schema/elevenlabs.go
+++ b/core/schema/elevenlabs.go
@ -4,3 +4,11 @@ type ElevenLabsTTSRequest struct {
 	Text    string `json:"text" yaml:"text"`
 	ModelID string `json:"model_id" yaml:"model_id"`
 }
+
+type ElevenLabsSoundGenerationRequest struct {
+	Text        string   `json:"text" yaml:"text"`
+	ModelID     string   `json:"model_id" yaml:"model_id"`
+	Duration    *float32 `json:"duration_seconds,omitempty" yaml:"duration_seconds,omitempty"`
+	Temperature *float32 `json:"prompt_influence,omitempty" yaml:"prompt_influence,omitempty"`
+	DoSample    *bool    `json:"do_sample,omitempty" yaml:"do_sample,omitempty"`
+}