Melvin Strobl
16383ea6e2
- `stt`: one realtime transcription session per utterance on the speech
server's url, opened at `listening` true or the first chunk, committed at
false; a whole clip goes to /v1/audio/transcriptions. Returns at once and
publishes `partial` and `text` when the server answers.
- `tts`: one /v1/audio/speech call per sentence, cached by voice and text in
node state; a reference is marked `append` while earlier ones still play.
- wren: `reply` says what was applied, `unhandled` carries the prompt when
nothing was. The install hint names the git pin (PyPI's `wren` is another
package).
- ai/client: transcribe, speak and transcribe_stream beside stream_completion.
- A device socket publishes scalar `in` ports as {"type": "publish"} text,
in order with its binary frames.
- Pipeline.publish walks only what its values reach: a node downstream of a
consumer that published nothing no longer runs again (wren re-acted on
every microphone chunk).
271 lines
9.3 KiB
Python
271 lines
9.3 KiB
Python
"""One OpenAI-compatible provider call, and the model listing beside it.
|
|
|
|
Chat completions with function tools is the dialect every hosted provider and
|
|
every local server speaks — OpenAI, Anthropic's compatibility endpoint,
|
|
OpenRouter, Ollama, vLLM — so one client covers them without a provider SDK.
|
|
The admin types the URL; the model never chooses where a request goes.
|
|
|
|
The call is streamed, so the answer reaches the browser as it is written.
|
|
|
|
Speech is the same dialect's audio half: a realtime transcription session, a
|
|
whole clip transcribed, and a sentence spoken, as `parrot serve` answers them.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import base64
|
|
import json
|
|
from collections.abc import AsyncIterator
|
|
from typing import Any
|
|
from urllib.parse import urlparse
|
|
|
|
import httpx
|
|
from websockets.asyncio.client import connect
|
|
|
|
|
|
class ProviderError(Exception):
|
|
"""The provider answered, and the answer was a refusal."""
|
|
|
|
|
|
def check_base(api_base: str) -> None:
|
|
parsed = urlparse(api_base)
|
|
if parsed.scheme not in ("http", "https") or not parsed.hostname:
|
|
raise ValueError("The provider URL must be an http(s) URL")
|
|
|
|
|
|
def _headers(api_key: str) -> dict[str, str]:
|
|
return {"Authorization": f"Bearer {api_key}"}
|
|
|
|
|
|
def _detail(response: httpx.Response) -> str:
|
|
try:
|
|
error = response.json().get("error")
|
|
except ValueError:
|
|
error = None
|
|
if isinstance(error, dict):
|
|
error = error.get("message")
|
|
text = str(error or response.text or response.reason_phrase)
|
|
return f"{response.status_code}: {text[:300]}"
|
|
|
|
|
|
class Assembler:
|
|
"""The streamed chunks put back together into the message a plain call returns.
|
|
|
|
The providers stream the same ``data:`` lines and differ only in the
|
|
margins: OpenRouter interleaves ``:`` comment lines, a thinking model's
|
|
reasoning arrives in a field of its own and is left out, OpenAI and
|
|
Anthropic send a tool call's arguments in pieces where Ollama sends the call
|
|
whole, and some leave out the call's ``index``.
|
|
"""
|
|
|
|
def __init__(self) -> None:
|
|
self.content = ""
|
|
self.calls: list[dict[str, Any]] = []
|
|
|
|
def feed(self, line: str) -> str:
|
|
"""One line of the stream; returns the text it adds to the answer."""
|
|
if not line.startswith("data:"):
|
|
return ""
|
|
data = line[5:].strip()
|
|
if not data or data == "[DONE]":
|
|
return ""
|
|
try:
|
|
chunk = json.loads(data)
|
|
except ValueError as exc:
|
|
raise ProviderError(f"sent {data[:300]}") from exc
|
|
error = chunk.get("error")
|
|
if error:
|
|
if isinstance(error, dict):
|
|
error = error.get("message", error)
|
|
raise ProviderError(str(error)[:300])
|
|
choices = chunk.get("choices") or [{}]
|
|
delta = choices[0].get("delta") or {}
|
|
for piece in delta.get("tool_calls") or []:
|
|
self._merge(piece)
|
|
text: str = delta.get("content") or ""
|
|
self.content += text
|
|
return text
|
|
|
|
def _merge(self, piece: dict[str, Any]) -> None:
|
|
index = piece.get("index", len(self.calls))
|
|
while len(self.calls) <= index:
|
|
self.calls.append(
|
|
{
|
|
"id": "",
|
|
"type": "function",
|
|
"function": {"name": "", "arguments": ""},
|
|
}
|
|
)
|
|
call = self.calls[index]
|
|
call["id"] = piece.get("id") or call["id"]
|
|
function = piece.get("function") or {}
|
|
call["function"]["name"] = function.get("name") or call["function"]["name"]
|
|
call["function"]["arguments"] += function.get("arguments") or ""
|
|
|
|
def message(self) -> dict[str, Any]:
|
|
return {
|
|
"role": "assistant",
|
|
"content": self.content or None,
|
|
"tool_calls": self.calls or None,
|
|
}
|
|
|
|
|
|
async def stream_completion(
|
|
*,
|
|
api_base: str,
|
|
api_key: str,
|
|
model: str,
|
|
messages: list[dict[str, Any]],
|
|
tools: list[dict[str, Any]],
|
|
timeout: float = 120.0,
|
|
) -> AsyncIterator[str | dict[str, Any]]:
|
|
"""``POST {api_base}/chat/completions`` streamed.
|
|
|
|
Yields each piece of the answer's text as it arrives, then the whole
|
|
message. The timeout is per chunk, so a long answer is not cut off.
|
|
"""
|
|
body: dict[str, Any] = {"model": model, "messages": messages, "stream": True}
|
|
if tools:
|
|
body["tools"] = tools
|
|
assembler = Assembler()
|
|
async with (
|
|
httpx.AsyncClient(timeout=timeout) as http,
|
|
http.stream(
|
|
"POST",
|
|
api_base.rstrip("/") + "/chat/completions",
|
|
json=body,
|
|
headers=_headers(api_key),
|
|
) as response,
|
|
):
|
|
if response.is_error:
|
|
await response.aread()
|
|
raise ProviderError(_detail(response))
|
|
async for line in response.aiter_lines():
|
|
text = assembler.feed(line)
|
|
if text:
|
|
yield text
|
|
yield assembler.message()
|
|
|
|
|
|
async def list_models(api_base: str, api_key: str, timeout: float = 15.0) -> list[str]:
|
|
"""The ids ``GET {api_base}/models`` offers, for the settings screen."""
|
|
async with httpx.AsyncClient(timeout=timeout) as http:
|
|
response = await http.get(
|
|
api_base.rstrip("/") + "/models", headers=_headers(api_key)
|
|
)
|
|
if response.is_error:
|
|
raise ProviderError(_detail(response))
|
|
data = response.json().get("data") or []
|
|
return sorted(str(m["id"]) for m in data if isinstance(m, dict) and "id" in m)
|
|
|
|
|
|
async def transcribe(
|
|
api_base: str,
|
|
audio: bytes,
|
|
filename: str,
|
|
media_type: str,
|
|
*,
|
|
model: str = "whisper-1",
|
|
timeout: float = 30.0,
|
|
) -> str:
|
|
"""``POST {api_base}/audio/transcriptions``: a whole clip, and what it says."""
|
|
async with httpx.AsyncClient(timeout=timeout) as http:
|
|
response = await http.post(
|
|
api_base.rstrip("/") + "/audio/transcriptions",
|
|
files={"file": (filename, audio, media_type)},
|
|
data={"model": model, "response_format": "json"},
|
|
)
|
|
if response.is_error:
|
|
raise ProviderError(_detail(response))
|
|
return str(response.json().get("text") or "")
|
|
|
|
|
|
async def speak(
|
|
api_base: str,
|
|
text: str,
|
|
voice: str,
|
|
*,
|
|
model: str = "tts-1",
|
|
timeout: float = 30.0,
|
|
) -> bytes:
|
|
"""``POST {api_base}/audio/speech``: ``text`` spoken, as WAV."""
|
|
async with httpx.AsyncClient(timeout=timeout) as http:
|
|
response = await http.post(
|
|
api_base.rstrip("/") + "/audio/speech",
|
|
json={
|
|
"model": model,
|
|
"input": text,
|
|
"voice": voice,
|
|
"response_format": "wav",
|
|
},
|
|
)
|
|
if response.is_error:
|
|
raise ProviderError(_detail(response))
|
|
return response.content
|
|
|
|
|
|
async def transcribe_stream(
|
|
api_base: str,
|
|
audio: asyncio.Queue[bytes | None],
|
|
*,
|
|
rate: int = 16000,
|
|
idle_s: float = 10.0,
|
|
) -> AsyncIterator[tuple[bool, str]]:
|
|
"""``WS {api_base}/realtime``: one utterance, transcribed while it is spoken.
|
|
|
|
Sends each chunk of PCM16 taken off ``audio`` and commits at ``None``.
|
|
Yields ``(False, delta)`` as words are agreed on, then ``(True,
|
|
transcript)`` once. A chunk that takes longer than ``idle_s`` to come
|
|
raises ``TimeoutError``, so an utterance nobody commits ends.
|
|
"""
|
|
url = "ws" + api_base.rstrip("/").removeprefix("http") + "/realtime"
|
|
async with connect(url, open_timeout=idle_s) as ws:
|
|
update = {"type": "audio/pcm", "rate": rate}
|
|
await ws.send(
|
|
json.dumps(
|
|
{
|
|
"type": "session.update",
|
|
"session": {
|
|
"type": "transcription",
|
|
"audio": {"input": {"format": update}},
|
|
},
|
|
}
|
|
)
|
|
)
|
|
|
|
async def append() -> None:
|
|
try:
|
|
while (
|
|
chunk := await asyncio.wait_for(audio.get(), idle_s)
|
|
) is not None:
|
|
data = base64.b64encode(chunk).decode()
|
|
await ws.send(
|
|
json.dumps({"type": "input_audio_buffer.append", "audio": data})
|
|
)
|
|
await ws.send(json.dumps({"type": "input_audio_buffer.commit"}))
|
|
except BaseException:
|
|
# Ends the reading below, which then says why.
|
|
await ws.close()
|
|
raise
|
|
|
|
sender = asyncio.create_task(append())
|
|
sender.add_done_callback(lambda done: done.cancelled() or done.exception())
|
|
try:
|
|
async for raw in ws:
|
|
event = json.loads(raw)
|
|
kind = str(event.get("type") or "")
|
|
if kind == "error":
|
|
raise ProviderError(str((event.get("error") or {}).get("message")))
|
|
if kind.endswith("input_audio_transcription.delta"):
|
|
yield False, str(event.get("delta") or "")
|
|
elif kind.endswith("input_audio_transcription.completed"):
|
|
yield True, str(event.get("transcript") or "")
|
|
return
|
|
finally:
|
|
sender.cancel()
|
|
failure = (
|
|
sender.exception() if sender.done() and not sender.cancelled() else None
|
|
)
|
|
raise failure or ProviderError("the session closed before its transcript")
|