feat(tts): add pocket-tts engine alongside kokoro-fastapi
Some checks failed
Release / Test backend (push) Successful in 36s
Release / Check ui (push) Successful in 50s
Release / Docker / caddy (push) Failing after 45s
CI / Test backend (pull_request) Successful in 39s
CI / Check ui (pull_request) Failing after 11s
CI / Docker / ui (pull_request) Has been skipped
CI / Docker / caddy (pull_request) Failing after 11s
Release / Docker / backend (push) Failing after 48s
Release / Docker / runner (push) Failing after 58s
Release / Upload source maps (push) Failing after 39s
CI / Docker / backend (pull_request) Failing after 34s
CI / Docker / runner (pull_request) Failing after 30s
Release / Docker / ui (push) Successful in 2m42s
Release / Gitea Release (push) Has been skipped

- New backend/internal/pockettts package: POST /tts client for
  kyutai-labs/pocket-tts; streams WAV response and transcodes to MP3
  via ffmpeg so the rest of the pipeline stays format-agnostic
- config.PocketTTS struct + POCKET_TTS_URL env var
- runner.Dependencies.PocketTTS field; runAudioTask routes by voice name:
  pockettts.IsPocketTTSVoice (alba, marius, javert, …) → pocket-tts,
  everything else → kokoro-fastapi
- Dockerfile: runner stage switched from distroless/static to alpine:3.21
  with ffmpeg + ca-certificates installed
- homelab compose: runner gets POCKET_TTS_URL=http://pocket-tts:8000
- Doppler prd + prd_homelab: KOKORO_URL=https://tts.libnovel.cc,
  POCKET_TTS_URL=https://pocket-tts.libnovel.cc
This commit is contained in:
Admin
2026-03-27 16:17:13 +05:00
parent 4831c74acc
commit fea09e3e23
6 changed files with 223 additions and 13 deletions

View File

@@ -31,6 +31,7 @@ import (
"github.com/libnovel/backend/internal/kokoro"
"github.com/libnovel/backend/internal/meili"
"github.com/libnovel/backend/internal/orchestrator"
"github.com/libnovel/backend/internal/pockettts"
"github.com/libnovel/backend/internal/scraper"
"github.com/libnovel/backend/internal/taskqueue"
)
@@ -84,8 +85,11 @@ type Dependencies struct {
SearchIndex meili.Client
// Novel is the scraper implementation.
Novel scraper.NovelScraper
// Kokoro is the TTS client.
// Kokoro is the Kokoro-FastAPI TTS client (GPU, OpenAI-compatible voices).
Kokoro kokoro.Client
// PocketTTS is the pocket-tts client (CPU, kyutai voices: alba, marius, etc.).
// If nil, pocket-tts voice tasks will fail with a clear error.
PocketTTS pockettts.Client
// Log is the structured logger.
Log *slog.Logger
}
@@ -448,14 +452,31 @@ func (r *Runner) runAudioTask(ctx context.Context, task domain.AudioTask) {
return
}
if r.deps.Kokoro == nil {
fail("kokoro client not configured")
return
}
audioData, err := r.deps.Kokoro.GenerateAudio(ctx, text, task.Voice)
if err != nil {
fail(fmt.Sprintf("kokoro generate: %v", err))
return
var audioData []byte
if pockettts.IsPocketTTSVoice(task.Voice) {
if r.deps.PocketTTS == nil {
fail("pocket-tts client not configured (POCKET_TTS_URL is empty)")
return
}
var genErr error
audioData, genErr = r.deps.PocketTTS.GenerateAudio(ctx, text, task.Voice)
if genErr != nil {
fail(fmt.Sprintf("pocket-tts generate: %v", genErr))
return
}
log.Info("runner: audio generated via pocket-tts", "voice", task.Voice)
} else {
if r.deps.Kokoro == nil {
fail("kokoro client not configured (KOKORO_URL is empty)")
return
}
var genErr error
audioData, genErr = r.deps.Kokoro.GenerateAudio(ctx, text, task.Voice)
if genErr != nil {
fail(fmt.Sprintf("kokoro generate: %v", genErr))
return
}
log.Info("runner: audio generated via kokoro-fastapi", "voice", task.Voice)
}
key := r.deps.AudioStore.AudioObjectKey(task.Slug, task.Chapter, task.Voice)