@vaibhavgupta0786/opencode-voice 0.0.0-stage → 0.3.0

package/servers/mlx_server.py

@@ -0,0 +1,94 @@
+#!/usr/bin/env python3
+"""Minimal OpenAI-compatible speech-to-text server on Apple Silicon GPU.
+
+Same API as the faster-whisper reference server, so opencode-voice needs no
+changes: POST /v1/audio/transcriptions (multipart file + model) -> {"text"}.
+
+Differences: mlx-whisper on the Metal GPU (~2x faster than CPU
+faster-whisper/small on this class of machine) and zero temp files — the
+uploaded WAV is decoded in memory with the stdlib wave module.
+
+Config (env):
+    MLX_MODEL    HF repo (default: mlx-community/whisper-small-mlx)
+    VOICE_STT_PORT  listen port (default: 8080)
+"""
+
+import io
+import json
+import os
+import wave
+from email.parser import BytesParser
+from email.policy import default as email_policy
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+import numpy as np
+import mlx_whisper
+
+MODEL = os.environ.get("MLX_MODEL", "mlx-community/whisper-small-mlx")
+PORT = int(os.environ.get("VOICE_STT_PORT", "8080"))
+
+print(f"loading {MODEL} ...", flush=True)
+# Warm up by loading weights once; per-request calls reuse them.
+_MODEL_REF = MODEL
+print("ready", flush=True)
+
+
+def wav_bytes_to_float32(raw: bytes) -> np.ndarray:
+    with wave.open(io.BytesIO(raw), "rb") as w:
+        if w.getframerate() != 16_000 or w.getnchannels() != 1 or w.getsampwidth() != 2:
+            raise ValueError(
+                f"expected 16kHz mono s16, got {w.getframerate()}Hz "
+                f"{w.getnchannels()}ch {w.getsampwidth() * 8}bit"
+            )
+        frames = w.readframes(w.getnframes())
+    return np.frombuffer(frames, dtype=np.int16).astype(np.float32) / 32768.0
+
+
+class Handler(BaseHTTPRequestHandler):
+    def _json(self, code: int, payload: dict) -> None:
+        body = json.dumps(payload).encode()
+        self.send_response(code)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def do_GET(self):  # noqa: N802
+        if self.path == "/health":
+            self._json(200, {"status": "ok"})
+        else:
+            self._json(404, {"error": "not found"})
+
+    def do_POST(self):  # noqa: N802
+        if not self.path.rstrip("/").endswith("/audio/transcriptions"):
+            self._json(404, {"error": "not found"})
+            return
+        length = int(self.headers.get("Content-Length", 0))
+        raw = self.rfile.read(length)
+        ctype = self.headers.get("Content-Type", "")
+        try:
+            msg = BytesParser(policy=email_policy).parsebytes(
+                f"Content-Type: {ctype}\r\nMIME-Version: 1.0\r\n\r\n".encode() + raw
+            )
+            audio: bytes | None = None
+            for part in msg.iter_parts():
+                if part.get_filename():
+                    audio = part.get_content()
+                    break
+            if audio is None:
+                self._json(400, {"error": "no file part"})
+                return
+            samples = wav_bytes_to_float32(bytes(audio))
+            result = mlx_whisper.transcribe(samples, path_or_hf_repo=_MODEL_REF)
+            self._json(200, {"text": (result.get("text") or "").strip()})
+        except Exception as error:  # keep the endpoint fail-closed with a reason
+            self._json(500, {"error": f"{type(error).__name__}: {error}"})
+
+    def log_message(self, *args):  # quieter logs
+        pass
+
+
+if __name__ == "__main__":
+    server = ThreadingHTTPServer(("127.0.0.1", PORT), Handler)
+    print(f"listening on 127.0.0.1:{PORT} model={MODEL}", flush=True)
+    server.serve_forever()

package/skills/voice-check/SKILL.md

@@ -0,0 +1,68 @@
+---
+name: Voice Check
+description: Check and restart the local whisper STT server that the opencode-voice dictation plugin uses on port 8080. Use when voice dictation fails, an "STT failed" error appears, or the user asks to fix, prepare, check, or bring back voice input.
+slash: true
+---
+
+# Voice STT server check
+
+The opencode-voice plugin transcribes dictation through a local whisper
+server at http://127.0.0.1:8080. When that server is down, every take
+fails with "STT failed". Work these steps in order; stop at the first one
+that resolves the situation.
+
+## 1. Health check
+
+    curl -s -m 3 http://127.0.0.1:8080/health
+
+If it prints `{"status": "ok"}`, the server is up — report that dictation
+should work (press Fn+F9) and stop. Nothing is broken.
+
+## 2. Locate the server install
+
+Check these directories in order; the server dir is the first that
+contains a server file and a Python virtualenv:
+
+1. `~/.local/share/opencode/voice-stt` — canonical location (see the
+   opencode-voice repo's `docs/STT.md`)
+2. `~/.local/share/opencode/opencode-dictate` — legacy location
+
+Inside the server dir, expect one of:
+
+- `mlx_server.py` + `mlx-env/` — mlx-whisper on the Apple Silicon GPU
+  (preferred), or
+- `server.py` + `stt-env/` — faster-whisper on CPU
+
+If neither directory exists, the STT server was never installed. Point the
+user to the opencode-voice repo's `docs/STT.md` and stop.
+
+## 3. Restart the server
+
+Clear anything still holding the port (ignore errors):
+
+    lsof -ti:8080 | xargs kill -9
+
+Start detached. If `mlx_server.py` exists:
+
+    cd <server-dir> && nohup ./mlx-env/bin/python3 mlx_server.py > mlx-server.log 2>&1 &
+
+otherwise:
+
+    cd <server-dir> && WHISPER_MODEL=small nohup ./stt-env/bin/python3 server.py > stt-server.log 2>&1 &
+
+## 4. Wait for boot, then re-verify
+
+Poll the health endpoint every 5 seconds, up to 30 seconds total — the
+model loads from cache in roughly 10-25 s on a cold start:
+
+    curl -s -m 3 http://127.0.0.1:8080/health
+
+## 5. If it still will not start
+
+Show the last 10-20 lines of the server log (`mlx-server.log` or
+`stt-server.log`) and report the error you see. Do not guess further.
+
+## 6. Report
+
+One line: server was up / has been restarted / failed (with the error).
+Remind the user to press Fn+F9 to dictate.