Muse Voice Transcribe

You're Claude Code, working in my repository.

Task
Pick the Muse Voice Transcribe recipe on this page that fits my project (transcribing recordings and live speech with speaker turns, or exact voice commands), tell me which and why, then build its core: send a mono 16-bit WAV to POST /v1/asr/transcribe as a two-part multipart request and print the transcript and any speaker turns.

Read this first
Meta's docs page "Muse Voice Transcribe" as Markdown: https://dev.meta.ai/docs/cookbook/muse-voice-transcribe.md
It's the source of truth for request and response shapes. The same page for people: https://dev.meta.ai/docs/cookbook/muse-voice-transcribe

Setup
- Base URL: https://api.meta.ai/v1
- Auth: send Authorization: Bearer $MODEL_API_KEY, reading MODEL_API_KEY from the environment. Never hard-code, print or commit the key.
- Model: muse-voice-transcribe-1.0
- Python: the OpenAI SDK (pip install openai) with base_url="https://api.meta.ai/v1" and api_key=os.environ["MODEL_API_KEY"] works, or plain requests.

Facts to get right
- File transcription: POST https://api.meta.ai/v1/asr/transcribe, multipart with a request part (JSON with model muse-voice-transcribe-1.0, audioEncoding WAV, and optional mode, keywords, languageBias) and an audio part.
- Only mono 16-bit PCM WAV at 16 or 24 kHz is accepted; convert first (ffmpeg -i in.mp3 -ac 1 -ar 24000 -sample_fmt s16 out.wav).
- Modes: PUSH_TO_TALK (the default), ENDPOINTING, DIARIZATION. Speaker labels are bare letters ("A"); skip turns with an empty transcript.
- Live speech uses wss://api.meta.ai/v1/asr/realtime: the credential travels in the first (handshake) frame, and audio must be paced at real time.
- $0.18 per hour of audio, billed in whole seconds. One request takes up to 10 minutes or 32 MB of audio.

Starter code (Python)
import json
import os

import requests

with open("interview.wav", "rb") as audio:
    res = requests.post(
        "https://api.meta.ai/v1/asr/transcribe",
        headers={"Authorization": f"Bearer {os.environ['MODEL_API_KEY']}"},
        files={
            "request": (None, json.dumps({"model": "muse-voice-transcribe-1.0", "audioEncoding": "WAV", "mode": "DIARIZATION"})),
            "audio": ("interview.wav", audio, "audio/wav"),
        },
    )
res.raise_for_status()
reply = res.json()
print(reply["transcript"].strip())
for turn in reply.get("turns", []):
    if turn.get("transcript"):
        print(f"[{turn['startMs'] / 1000:7.2f}s] {turn.get('speaker', '?')}: {turn['transcript']}")

Done when
- A WAV file transcribes and the transcript prints.
- With mode DIARIZATION, the turns print with their speaker letters and start times.
- Other formats are converted to WAV first, and a 400's message is shown.
- The key comes from MODEL_API_KEY and never appears in code, logs or commits.
- API errors (4xx, 429) surface their message instead of failing silently.