@coralai/sps-cli 0.59.0 → 0.60.1

package/sidecar/kokoro-tts/README.md

@@ -1,31 +0,0 @@
-# Kokoro TTS 边车
-
-sps-cli 的语音输出引擎(替代浏览器 Web Speech)。本地推理、中文优先。
-
-- **模型**:Kokoro-82M v1.1-zh(StyleTTS2,ONNX),中文专业音色 `zf_*`/`zm_*`。
-- **g2p**:misaki `ZHG2P(version="1.1")`(jieba + pypinyin),中文多音字/分词准。
-- **形态**:纯 stdlib HTTP(`/healthz` + `POST /tts→wav`),不引 web 框架。
-
-## 生命周期由 sps-cli 托管
-
-不要手动跑本文件。装/起/停走:
-
-```bash
-sps kokoro setup     # 建 venv + pip 安装 + 下载模型(~300MB,首次)
-sps kokoro status    # 查状态
-sps kokoro stop      # 停边车
-```
-
-`sps console` 启动时会在已安装的前提下自动拉起,退出时停。前端 `/api/tts` 反代到本边车。
-
-## 手动自检
-
-```bash
-curl -s -X POST http://127.0.0.1:17650/tts \
-  -H 'content-type: application/json' \
-  -d '{"text":"你好,世界。今天天气不错。"}' -o /tmp/kokoro.wav
-```
-
-## 环境变量
-
-见 `server.py` 顶部 docstring(端口/模型路径/默认音色)。管家 `KokoroTtsManager.ts` 是这些值的单一来源。

package/sidecar/kokoro-tts/requirements.txt

@@ -1,4 +0,0 @@
-# Kokoro TTS 边车依赖(中文优先)。onnxruntime / numpy 由 kokoro-onnx 传递引入。
-# misaki[zh] 带来中文 g2p(jieba 分词 + pypinyin),这是中文音质的关键一环。
-kokoro-onnx>=0.4.9
-misaki[zh]>=0.9.4

package/sidecar/kokoro-tts/server.py

@@ -1,135 +0,0 @@
-"""Kokoro TTS 边车 —— sps-cli 的语音输出引擎(中文优先)。
-
-这是一个**纯 stdlib HTTP 服务**(除推理依赖 kokoro-onnx / misaki 外不引 web 框架):
-  - GET  /healthz   → 模型就绪后返回 200 "ok"(边车管家据此判活)
-  - POST /tts       → {"text": "...", "voice"?: "zf_001", "speed"?: 1.0} → audio/wav
-
-设计约束(见仓库 CLAUDE.md「红线」):这是 sps-cli 的**运行时执行 substrate**,只做
-一件事——文字→音素(misaki 中文 g2p)→ Kokoro 合成 → WAV 字节。所有生命周期(装/起/停/
-健康)由上层 KokoroTtsManager.ts 负责,本文件不管进程编排。
-
-环境变量:
-  SPS_KOKORO_PORT     监听端口(默认 17650,与管家单一来源保持一致)
-  SPS_KOKORO_MODEL    kokoro-v1.1-zh.onnx 路径
-  SPS_KOKORO_VOICES   voices-v1.1-zh.bin 路径
-  SPS_KOKORO_CONFIG   config.json 路径(中文 vocab)
-  SPS_KOKORO_VOICE    默认音色 id(默认 zf_001)
-"""
-
-from __future__ import annotations
-
-import io
-import json
-import os
-import sys
-import threading
-import wave
-from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
-
-import numpy as np
-from kokoro_onnx import Kokoro
-from kokoro_onnx.tokenizer import Tokenizer
-from misaki import zh
-
-PORT = int(os.environ.get("SPS_KOKORO_PORT", "17650"))
-MODEL = os.environ.get("SPS_KOKORO_MODEL", "kokoro-v1.1-zh.onnx")
-VOICES = os.environ.get("SPS_KOKORO_VOICES", "voices-v1.1-zh.bin")
-CONFIG = os.environ.get("SPS_KOKORO_CONFIG", "config.json")
-DEFAULT_VOICE = os.environ.get("SPS_KOKORO_VOICE", "zf_001")
-
-# 推理不保证线程安全(onnxruntime run 可并发,但 g2p 状态不一定),整体串行化。
-# 前端本就逐句发,串行足够,且避免竞态。
-_lock = threading.Lock()
-
-
-def _load() -> tuple[Kokoro, "zh.ZHG2P"]:
-    print(f"[kokoro] loading model={MODEL} voices={VOICES}", file=sys.stderr, flush=True)
-    kokoro = Kokoro(MODEL, VOICES, vocab_config=CONFIG)
-    # 中文 g2p 只处理中文;夹在里面的英文段必须挂英文回退,否则被丢成 unk(❓)不发音。
-    # 复用 kokoro-onnx 自带的 espeak 分词器(同库、音素表对齐、无额外依赖)。
-    en_tok = Tokenizer()
-    g2p = zh.ZHG2P(version="1.1", en_callable=lambda en: en_tok.phonemize(en, lang="en-us"))
-    print("[kokoro] ready", file=sys.stderr, flush=True)
-    return kokoro, g2p
-
-
-def _to_wav(samples: np.ndarray, sample_rate: int) -> bytes:
-    """float32 [-1,1] → 16-bit PCM WAV 字节。"""
-    pcm = np.clip(np.asarray(samples, dtype=np.float32), -1.0, 1.0)
-    pcm16 = (pcm * 32767.0).astype("<i2")
-    buf = io.BytesIO()
-    with wave.open(buf, "wb") as wav:
-        wav.setnchannels(1)
-        wav.setsampwidth(2)
-        wav.setframerate(int(sample_rate))
-        wav.writeframes(pcm16.tobytes())
-    return buf.getvalue()
-
-
-class Handler(BaseHTTPRequestHandler):
-    # 复用 server 实例上加载好的模型(见 main)。
-    kokoro: Kokoro
-    g2p: "zh.ZHG2P"
-
-    def log_message(self, *_args) -> None:  # 静音默认访问日志(边车日志由管家收集)
-        pass
-
-    def _send(self, code: int, body: bytes, content_type: str) -> None:
-        self.send_response(code)
-        self.send_header("Content-Type", content_type)
-        self.send_header("Content-Length", str(len(body)))
-        self.end_headers()
-        self.wfile.write(body)
-
-    def do_GET(self) -> None:
-        if self.path == "/healthz":
-            self._send(200, b"ok", "text/plain")
-        else:
-            self._send(404, b"not found", "text/plain")
-
-    def do_POST(self) -> None:
-        if self.path != "/tts":
-            self._send(404, b"not found", "text/plain")
-            return
-        try:
-            length = int(self.headers.get("Content-Length", "0"))
-            payload = json.loads(self.rfile.read(length) or b"{}")
-            text = (payload.get("text") or "").strip()
-            voice = payload.get("voice") or DEFAULT_VOICE
-            speed = float(payload.get("speed") or 1.0)
-        except (ValueError, json.JSONDecodeError) as err:
-            self._send(400, f"bad request: {err}".encode(), "text/plain")
-            return
-
-        if not text:
-            self._send(400, b"missing text", "text/plain")
-            return
-
-        try:
-            with _lock:
-                phonemes, _ = self.g2p(text)
-                samples, sample_rate = self.kokoro.create(
-                    phonemes, voice=voice, speed=speed, is_phonemes=True
-                )
-            self._send(200, _to_wav(samples, sample_rate), "audio/wav")
-        except Exception as err:  # noqa: BLE001 —— 边车稳定性优先,任何合成错误回 500 不崩进程
-            print(f"[kokoro] synth error: {err}", file=sys.stderr, flush=True)
-            self._send(500, f"synth failed: {err}".encode(), "text/plain")
-
-
-def main() -> None:
-    kokoro, g2p = _load()
-    Handler.kokoro = kokoro
-    Handler.g2p = g2p
-    server = ThreadingHTTPServer(("127.0.0.1", PORT), Handler)
-    print(f"[kokoro] listening on 127.0.0.1:{PORT}", file=sys.stderr, flush=True)
-    try:
-        server.serve_forever()
-    except KeyboardInterrupt:
-        pass
-    finally:
-        server.server_close()
-
-
-if __name__ == "__main__":
-    main()