Files
libnovel/backend/internal/pockettts/client.go
Admin a771405db8
Some checks failed
CI / Backend (push) Successful in 48s
CI / UI (push) Successful in 28s
Release / Check ui (push) Successful in 39s
Release / Test backend (push) Successful in 49s
Release / Docker / caddy (push) Successful in 59s
CI / Backend (pull_request) Successful in 41s
CI / UI (pull_request) Successful in 46s
Release / Docker / ui (push) Successful in 1m31s
Release / Docker / backend (push) Successful in 3m27s
Release / Docker / runner (push) Successful in 3m47s
Release / Gitea Release (push) Failing after 32s
feat(audio): WAV streaming, bulk audio generation admin endpoints, cancel/resume
- Add StreamAudioWAV() to pocket-tts and Kokoro clients; pocket-tts streams
  raw WAV directly (no ffmpeg), Kokoro requests response_format:wav with stream:true
- GET /api/audio-stream supports ?format=wav for lower-latency first-byte delivery;
  WAV cached separately in MinIO as {slug}/{n}/{voice}.wav
- Add GET /api/admin/audio/jobs with optional ?slug filter
- Add POST /api/admin/audio/bulk {slug, voice, from, to, skip_existing, force}
  where skip_existing=true (default) resumes interrupted bulk jobs
- Add POST /api/admin/audio/cancel-bulk {slug} to cancel all pending/running tasks
- Add CancelAudioTasksBySlug to taskqueue.Producer + asynqqueue implementation
- Add AudioObjectKeyExt to bookstore.AudioStore for format-aware MinIO keys

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-02 16:19:14 +05:00

255 lines
8.1 KiB
Go

// Package pockettts provides a client for the kyutai-labs/pocket-tts TTS service.
//
// pocket-tts exposes a non-OpenAI API:
//
// POST /tts (multipart form: text, voice_url) → streaming WAV
// GET /health → {"status":"healthy"}
//
// GenerateAudio streams the WAV response and transcodes it to MP3 using ffmpeg,
// so callers receive MP3 bytes — the same format as the kokoro client — and the
// rest of the pipeline does not need to care which TTS engine was used.
//
// StreamAudioMP3 is the streaming variant: it returns an io.ReadCloser that
// yields MP3-encoded audio incrementally as pocket-tts generates it, without
// buffering the full output.
//
// Predefined voices (pass the bare name as the voice parameter):
//
// alba, marius, javert, jean, fantine, cosette, eponine, azelma,
// anna, vera, charles, paul, george, mary, jane, michael, eve,
// bill_boerst, peter_yearsley, stuart_bell
package pockettts
import (
"bytes"
"context"
"fmt"
"io"
"mime/multipart"
"net/http"
"os/exec"
"strings"
"time"
)
// PredefinedVoices is the set of voice names built into pocket-tts.
// The runner uses this to decide which TTS engine to route a task to.
var PredefinedVoices = map[string]struct{}{
"alba": {}, "marius": {}, "javert": {}, "jean": {},
"fantine": {}, "cosette": {}, "eponine": {}, "azelma": {},
"anna": {}, "vera": {}, "charles": {}, "paul": {},
"george": {}, "mary": {}, "jane": {}, "michael": {},
"eve": {}, "bill_boerst": {}, "peter_yearsley": {}, "stuart_bell": {},
}
// IsPocketTTSVoice reports whether voice is served by pocket-tts.
func IsPocketTTSVoice(voice string) bool {
_, ok := PredefinedVoices[voice]
return ok
}
// Client is the interface for interacting with the pocket-tts service.
type Client interface {
// GenerateAudio synthesises text using the given voice and returns MP3 bytes.
// Voice must be one of the predefined pocket-tts voice names.
GenerateAudio(ctx context.Context, text, voice string) ([]byte, error)
// StreamAudioMP3 synthesises text and returns an io.ReadCloser that streams
// MP3-encoded audio incrementally via a live ffmpeg transcode pipe.
// The caller must always close the returned ReadCloser.
StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error)
// StreamAudioWAV synthesises text and returns an io.ReadCloser that streams
// raw WAV audio directly from pocket-tts without any transcoding.
// The stream begins with a WAV header followed by 16-bit PCM frames at 16 kHz.
// The caller must always close the returned ReadCloser.
StreamAudioWAV(ctx context.Context, text, voice string) (io.ReadCloser, error)
// ListVoices returns the available predefined voice names.
ListVoices(ctx context.Context) ([]string, error)
}
// httpClient is the concrete pocket-tts HTTP client.
type httpClient struct {
baseURL string
http *http.Client
}
// New returns a Client targeting baseURL (e.g. "https://pocket-tts.libnovel.cc").
func New(baseURL string) Client {
return &httpClient{
baseURL: strings.TrimRight(baseURL, "/"),
http: &http.Client{Timeout: 10 * time.Minute},
}
}
// GenerateAudio posts to POST /tts and transcodes the WAV response to MP3
// using the system ffmpeg binary. Requires ffmpeg to be on PATH (available in
// the runner Docker image via Alpine's ffmpeg package).
func (c *httpClient) GenerateAudio(ctx context.Context, text, voice string) ([]byte, error) {
if text == "" {
return nil, fmt.Errorf("pockettts: empty text")
}
if voice == "" {
voice = "alba"
}
resp, err := c.postTTS(ctx, text, voice)
if err != nil {
return nil, err
}
defer resp.Body.Close()
wavData, err := io.ReadAll(resp.Body)
if err != nil {
return nil, fmt.Errorf("pockettts: read response body: %w", err)
}
// ── Transcode WAV → MP3 via ffmpeg ────────────────────────────────────────
mp3Data, err := wavToMP3(ctx, wavData)
if err != nil {
return nil, fmt.Errorf("pockettts: transcode to mp3: %w", err)
}
return mp3Data, nil
}
// StreamAudioMP3 posts to POST /tts and returns an io.ReadCloser that delivers
// MP3 bytes as pocket-tts generates WAV frames. ffmpeg runs as a subprocess
// with stdin connected to the live WAV stream and stdout piped to the caller.
// The caller must always close the returned ReadCloser.
func (c *httpClient) StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error) {
if text == "" {
return nil, fmt.Errorf("pockettts: empty text")
}
if voice == "" {
voice = "alba"
}
resp, err := c.postTTS(ctx, text, voice)
if err != nil {
return nil, err
}
// Start ffmpeg: read WAV from stdin (the live HTTP body), write MP3 to stdout.
cmd := exec.CommandContext(ctx,
"ffmpeg",
"-hide_banner", "-loglevel", "error",
"-i", "pipe:0", // WAV from stdin
"-f", "mp3", // output format
"-q:a", "2", // VBR ~190 kbps
"pipe:1", // MP3 to stdout
)
cmd.Stdin = resp.Body
pr, pw := io.Pipe()
cmd.Stdout = pw
var stderrBuf bytes.Buffer
cmd.Stderr = &stderrBuf
if err := cmd.Start(); err != nil {
resp.Body.Close()
return nil, fmt.Errorf("pockettts: start ffmpeg: %w", err)
}
// Close the write end of the pipe when ffmpeg exits, propagating any error.
go func() {
waitErr := cmd.Wait()
resp.Body.Close()
if waitErr != nil {
pw.CloseWithError(fmt.Errorf("ffmpeg: %w (stderr: %s)", waitErr, stderrBuf.String()))
} else {
pw.Close()
}
}()
return pr, nil
}
// StreamAudioWAV posts to POST /tts and returns an io.ReadCloser that delivers
// raw WAV bytes directly from pocket-tts — no ffmpeg transcoding required.
// The first bytes will be a WAV header (RIFF/fmt chunk) followed by PCM frames.
// The caller must always close the returned ReadCloser.
func (c *httpClient) StreamAudioWAV(ctx context.Context, text, voice string) (io.ReadCloser, error) {
if text == "" {
return nil, fmt.Errorf("pockettts: empty text")
}
if voice == "" {
voice = "alba"
}
resp, err := c.postTTS(ctx, text, voice)
if err != nil {
return nil, err
}
return resp.Body, nil
}
// ListVoices returns the statically known predefined voice names.
// pocket-tts has no REST endpoint for listing voices.
func (c *httpClient) ListVoices(_ context.Context) ([]string, error) {
voices := make([]string, 0, len(PredefinedVoices))
for v := range PredefinedVoices {
voices = append(voices, v)
}
return voices, nil
}
// postTTS sends a multipart POST /tts request and returns the raw response.
// The caller is responsible for closing resp.Body.
func (c *httpClient) postTTS(ctx context.Context, text, voice string) (*http.Response, error) {
var body bytes.Buffer
mw := multipart.NewWriter(&body)
if err := mw.WriteField("text", text); err != nil {
return nil, fmt.Errorf("pockettts: write text field: %w", err)
}
if err := mw.WriteField("voice_url", voice); err != nil {
return nil, fmt.Errorf("pockettts: write voice_url field: %w", err)
}
if err := mw.Close(); err != nil {
return nil, fmt.Errorf("pockettts: close multipart writer: %w", err)
}
req, err := http.NewRequestWithContext(ctx, http.MethodPost,
c.baseURL+"/tts", &body)
if err != nil {
return nil, fmt.Errorf("pockettts: build request: %w", err)
}
req.Header.Set("Content-Type", mw.FormDataContentType())
resp, err := c.http.Do(req)
if err != nil {
return nil, fmt.Errorf("pockettts: request: %w", err)
}
if resp.StatusCode != http.StatusOK {
_, _ = io.Copy(io.Discard, resp.Body)
resp.Body.Close()
return nil, fmt.Errorf("pockettts: server returned %d", resp.StatusCode)
}
return resp, nil
}
// wavToMP3 converts raw WAV bytes to MP3 using ffmpeg.
// ffmpeg reads from stdin (pipe:0) and writes to stdout (pipe:1).
func wavToMP3(ctx context.Context, wav []byte) ([]byte, error) {
cmd := exec.CommandContext(ctx,
"ffmpeg",
"-hide_banner", "-loglevel", "error",
"-i", "pipe:0", // read WAV from stdin
"-f", "mp3", // output format
"-q:a", "2", // VBR quality ~190 kbps
"pipe:1", // write MP3 to stdout
)
cmd.Stdin = bytes.NewReader(wav)
var out, stderr bytes.Buffer
cmd.Stdout = &out
cmd.Stderr = &stderr
if err := cmd.Run(); err != nil {
return nil, fmt.Errorf("ffmpeg: %w (stderr: %s)", err, stderr.String())
}
return out.Bytes(), nil
}