stacks/index-tts: own FastAPI wrapper for IndexTTS-2 + deploy playbook

Adds a third TTS to the irv-ml1 fleet. IndexTTS-2 is Bilibili's emotion-controllable zero-shot TTS (paper 2506.21619). Distinguishing capability vs the existing two: timbre and emotion are disentangled — clone a voice's timbre from one reference and the emotion from a different reference, OR set emotion via 8-vector, OR derive it from a text description. Neither CosyVoice 3 nor Qwen3-TTS-1.7B-Base does this cleanly in English. Wrapper is owned end-to-end (~150 lines in app.py) — the only existing FastAPI fork (csllpr/index-tts-fastapi) targets v1 and is a dormant single-commit repo. Upstream IndexTTS-2 ships only a Gradio webui. Layout follows the qwen3-tts pattern: stacks/index-tts/ Dockerfile — CUDA 12.8 base, IndexTTS pinned to a SHA app.py — FastAPI: POST /v1/audio/speech + /v1/voices entrypoint.sh — one-time HF snapshot_download of the weights compose.yaml — env-driven, GPU pinning support, bind mounts .env.example — port 8192, fp16, paths README.md — API examples + comparison vs the other TTS playbooks/deploy-index-tts.yaml — elway playbook for irv-ml1 Voice and emotion libraries are flat host dirs of WAVs, bind-mounted. Drop a new <name>.wav and /v1/voices picks it up immediately. License caveat: IndexTTS-2 weights ship under a custom Bilibili license (free at our scale, not OSI-open). README documents it.
2026-04-25 10:38:14 -07:00
parent 1f14c6d959
commit b75f020cc9
7 changed files with 604 additions and 0 deletions
@@ -0,0 +1,53 @@
+# IndexTTS-2 stack tunables. Copy to `.env` on irv-ml1 before deploying.
+
+# ── build pin ────────────────────────────────────────────────────────
+# SHA of index-tts/index-tts to build from. Bump + rebuild when you want
+# upstream wrapper updates.
+INDEX_TTS_SHA=830f6f8f94a51fea23ab1d639027a86200075a4e
+
+# Local image tag — bump when you change build context (Dockerfile,
+# app.py, entrypoint.sh) to force a fresh layer build.
+INDEX_TTS_TAG=v1
+
+# ── network ──────────────────────────────────────────────────────────
+# Host port. Container listens on 8000 internally.
+# Reserved on irv-ml1: 8188 ComfyUI, 8190 CosyVoice, 8191 Qwen3-TTS,
+# 8765 Parakeet. 8192 is open.
+INDEX_TTS_PORT=8192
+
+# Bind address. 0.0.0.0 exposes on all interfaces (incl. WG tunnel
+# interface 10.100.79.3); 127.0.0.1 restricts to local-only.
+INDEX_TTS_BIND=0.0.0.0
+
+# ── runtime / GPU ────────────────────────────────────────────────────
+# bf16/fp16 vs fp32. Empty = fp32. "1" = fp16. IndexTTS-2 README
+# recommends fp16; ~6-10 GB VRAM at fp16, double at fp32.
+INDEX_TTS_FP16=1
+
+# GPU pinning. Empty = let IndexTTS auto-pick (cuda:0). Set to "cuda:1"
+# to pin to the second GPU (irv-ml1's RTX 3090 vs RTX A6000).
+INDEX_TTS_DEVICE=
+
+# Devices visible inside the container. Either "all" (both GPUs) or a
+# comma-separated list of indices (e.g. "1" to expose only the second
+# card). The DEVICE setting above further restricts within those.
+INDEX_TTS_GPU_DEVICES=all
+
+# Logging level for the wrapper itself (IndexTTS internals are noisier
+# regardless).
+INDEX_TTS_LOG_LEVEL=INFO
+
+# ── persistent storage on the host ───────────────────────────────────
+# Model weights (~5-7 GB after first run). Bind-mounted so model state
+# survives container recreate. Excluded from restic (regenerable from HF).
+INDEX_TTS_CACHE_DIR=/worktank/index-tts/cache
+
+# Voice library — flat dir of <name>.wav files (timbre references).
+# Cloned voices need the original reference audio to recreate; included
+# in restic.
+INDEX_TTS_VOICES_DIR=/worktank/index-tts/voices
+
+# Emotion library — flat dir of <name>.wav files (emotion references,
+# typically short clips with strong affect). Optional — without any
+# entries here you can still use emotion_vector or emotion_text.
+INDEX_TTS_EMOTIONS_DIR=/worktank/index-tts/emotions