Panniantong/Agent-Reach · error · TranscribeError

audio produced {len(chunks)} chunks; safety limit is {MAX_CH

Error message

audio produced {len(chunks)} chunks; safety limit is {MAX_CHUNKS} (~{max_minutes} minutes)

What it means

Raised after download/compress/chunk when the audio splits into more than MAX_CHUNKS (24) chunks of 10 minutes each, i.e. roughly 4 hours of audio. This is a cost and time safety valve so a single call cannot transcribe arbitrarily long media. It fires before any chunk is uploaded, so no API spend occurs.

Source

Thrown at agent_reach/transcribe.py:464

    work_dir.mkdir(parents=True, exist_ok=True)

    src_path = Path(source)
    if src_path.is_file():
        audio = src_path
    else:
        audio = download_audio(source, work_dir)

    _require_size_at_most(audio, MAX_SOURCE_BYTES, "source")
    _require_duration_within_budget(audio)
    compressed = compress_audio(audio, work_dir)
    if compressed.stat().st_size <= SIZE_LIMIT_BYTES:
        chunks = [compressed]
    else:
        chunks = chunk_audio(compressed, work_dir)

    if len(chunks) > MAX_CHUNKS:
        max_minutes = MAX_CHUNKS * CHUNK_SECONDS // 60
        raise TranscribeError(
            f"audio produced {len(chunks)} chunks; safety limit is "
            f"{MAX_CHUNKS} (~{max_minutes} minutes)"
        )
    chunk_sizes = [
        _require_size_at_most(chunk, SIZE_LIMIT_BYTES, f"chunk {chunk.name}")
        for chunk in chunks
    ]
    total_chunk_bytes = sum(chunk_sizes)
    if total_chunk_bytes > MAX_TOTAL_CHUNK_BYTES:
        limit_mib = MAX_TOTAL_CHUNK_BYTES / (1024 * 1024)
        raise TranscribeError(
            f"audio chunks total {total_chunk_bytes} bytes; "
            f"safety limit is {limit_mib:g} MiB"
        )

    pieces: List[str] = []
    for chunk in chunks:
        text = _transcribe_with_fallback(chunk, order, cfg)

View on GitHub (pinned to 93ae1d18c3)

Solutions

1. Trim or split the source externally (e.g. yt-dlp --download-sections or ffmpeg -ss/-t) and transcribe segments separately
2. Point transcribe() at a shorter portion of the media instead of the full item
3. If you own the deployment and accept the cost, raise MAX_CHUNKS in a fork — but prefer external splitting
4. For xiaoyuzhou episodes, use the bundled scripts/transcribe_xiaoyuzhou.sh flow which handles episode lengths itself

Example fix

# before
text = transcribe("https://www.youtube.com/watch?v=LONG_LIVESTREAM")  # 25 chunks > 24

# after — transcribe only the first 3 hours
# yt-dlp --download-sections "*0-3:00:00" -o seg.mp4 "<url>"
text = transcribe("seg.mp4")

from agent_reach.transcribe import MAX_AUDIO_SECONDS, MAX_CHUNKS, CHUNK_SECONDS

def duration_ok(seconds: float) -> bool:
    return seconds <= MAX_AUDIO_SECONDS  # 24 * 600 = 14400s

from agent_reach.transcribe import transcribe, TranscribeError
try:
    text = transcribe(url)
except TranscribeError as e:
    if "safety limit is" in str(e) and "chunks" in str(e):
        text = transcribe_in_parts(url, parts=4)  # your own splitter
    else:
        raise

When it happens

Trigger: Calling transcribe() on a source longer than MAX_AUDIO_SECONDS (24 * 600s = 14400s = 4 hours), or on audio that compresses poorly (e.g. already-compressed formats) so chunk_audio must split the oversized compressed file into many size-based segments exceeding 24.

Common situations: Transcribing long podcasts, livestream VODs, multi-hour talks, or audiobooks; feeding a high-bitrate source file where compression cannot get under SIZE_LIMIT_BYTES per 10-minute segment.

Related errors

• audio chunks total {total_chunk_bytes} bytes; safety limit i
• chunk segment duration must be positive
• ffmpeg produced no chunks
• audio duration exceeds safety limit of {max_minutes} minutes
• chunk generation safety limit is {MAX_CHUNKS}; segment durat

AI-assisted analysis of Panniantong/Agent-Reach@93ae1d18c3 (2026-08-14). Data as JSON: /api/errors/9d97534af74d06af. Report an issue: GitHub.