Files
ai-operator/internal/ai/tts/elevenlabs_ws_streaming.go
T

169 lines
3.7 KiB
Go

package tts
import (
"context"
"encoding/base64"
"encoding/json"
"errors"
"net/http"
"net/url"
"strconv"
"strings"
"sync"
"time"
"ai-operator/internal/config"
"github.com/gorilla/websocket"
)
type ElevenLabsWS struct {
cfg config.Config
conn *websocket.Conn
audio chan AudioChunk
mu sync.Mutex
closed bool
}
func NewElevenLabsWS(cfg config.Config) *ElevenLabsWS {
return &ElevenLabsWS{cfg: cfg, audio: make(chan AudioChunk, 64)}
}
func (p *ElevenLabsWS) Start(ctx context.Context, req TTSStreamRequest) error {
if p.cfg.Eleven.APIKey == "" {
return errors.New("ELEVENLABS_API_KEY is required for ElevenLabs streaming TTS")
}
if req.VoiceID == "" {
return errors.New("ElevenLabs voice id is required for streaming TTS")
}
base := strings.TrimRight(p.cfg.Eleven.TTSURL, "/") + "/" + url.PathEscape(req.VoiceID) + "/stream-input"
u, err := url.Parse(base)
if err != nil {
return err
}
q := u.Query()
q.Set("model_id", nonEmpty(req.ModelID, p.cfg.Eleven.TTSModelID))
q.Set("output_format", nonEmpty(req.OutputFormat, p.cfg.Eleven.TTSOutputFormat))
q.Set("optimize_streaming_latency", strconv.Itoa(p.cfg.Eleven.TTSOptimizeStreamingLatency))
u.RawQuery = q.Encode()
h := http.Header{}
h.Set("xi-api-key", p.cfg.Eleven.APIKey)
d := websocket.Dialer{HandshakeTimeout: p.cfg.Eleven.TTSTimeout}
conn, _, err := d.DialContext(ctx, u.String(), h)
if err != nil {
return err
}
p.mu.Lock()
p.conn = conn
p.audio = make(chan AudioChunk, 64)
p.closed = false
p.mu.Unlock()
bos := map[string]any{
"text": " ",
"voice_settings": map[string]any{
"stability": p.cfg.Eleven.TTSStability,
"similarity_boost": p.cfg.Eleven.TTSSimilarityBoost,
"style": p.cfg.Eleven.TTSStyle,
"use_speaker_boost": p.cfg.Eleven.TTSUseSpeakerBoost,
},
}
if err := conn.WriteJSON(bos); err != nil {
return err
}
go p.readLoop()
return nil
}
func (p *ElevenLabsWS) SendText(ctx context.Context, text string, flush bool) error {
p.mu.Lock()
conn := p.conn
p.mu.Unlock()
if conn == nil {
return errors.New("elevenlabs tts websocket not connected")
}
msg := map[string]any{"text": text, "try_trigger_generation": flush}
if flush {
msg["flush"] = true
}
return conn.WriteJSON(msg)
}
func (p *ElevenLabsWS) Audio() <-chan AudioChunk { return p.audio }
func (p *ElevenLabsWS) Close(ctx context.Context) error {
p.mu.Lock()
conn := p.conn
p.conn = nil
p.mu.Unlock()
if conn != nil {
_ = conn.WriteJSON(map[string]any{"text": ""})
_ = conn.WriteControl(websocket.CloseMessage, websocket.FormatCloseMessage(websocket.CloseNormalClosure, ""), time.Now().Add(time.Second))
_ = conn.Close()
}
p.closeAudio()
return nil
}
func (p *ElevenLabsWS) readLoop() {
for {
p.mu.Lock()
conn := p.conn
p.mu.Unlock()
if conn == nil {
p.closeAudio()
return
}
_, data, err := conn.ReadMessage()
if err != nil {
p.closeAudio()
return
}
audio := parseAudio(data)
if len(audio) == 0 {
continue
}
p.emit(AudioChunk{Data: audio, Timestamp: time.Now().UTC()})
}
}
func parseAudio(data []byte) []byte {
var m map[string]any
if json.Unmarshal(data, &m) == nil {
for _, k := range []string{"audio", "audio_base64"} {
if s, ok := m[k].(string); ok && s != "" {
b, _ := base64.StdEncoding.DecodeString(s)
return b
}
}
return nil
}
return data
}
func (p *ElevenLabsWS) emit(ch AudioChunk) {
p.mu.Lock()
defer p.mu.Unlock()
if p.closed {
return
}
select {
case p.audio <- ch:
default:
}
}
func (p *ElevenLabsWS) closeAudio() {
p.mu.Lock()
defer p.mu.Unlock()
if !p.closed {
close(p.audio)
p.closed = true
}
}
func nonEmpty(v, fallback string) string {
if v != "" {
return v
}
return fallback
}