run-llama/llama_index · error · ValueError

resolve_audio returned zero bytes

Error message

resolve_audio returned zero bytes

What it means

Raised by AudioBlock.resolve_audio when the resolved buffer (from raw bytes, path, or URL) has zero bytes. Same guard pattern as ImageBlock.resolve_image: it prevents empty audio payloads from reaching provider APIs.

Source

Thrown at llama-index-core/llama_index/core/base/llms/types.py:505

        """
        data_buffer = (
            self.audio
            if isinstance(self.audio, IOBase)
            else resolve_binary(
                raw_bytes=self.audio,
                path=self.path,
                url=str(self.url) if self.url else None,
                as_base64=as_base64,
            )
        )
        # Check size by seeking to end and getting position
        data_buffer.seek(0, 2)  # Seek to end
        size = data_buffer.tell()
        data_buffer.seek(0)  # Reset to beginning

        if size == 0:
            raise ValueError("resolve_audio returned zero bytes")
        return data_buffer

    def inline_url(self) -> str:
        b64 = self.resolve_audio(as_base64=True)
        b64_str = b64.read().decode("utf-8")
        if self.format:
            mimetype = filetype.get_type(ext=self.format).mime
            if mimetype:
                return f"data:{mimetype};base64,{b64_str}"
        return f"data:audio;base64,{b64_str}"

    async def aestimate_tokens(self, *args: Any, **kwargs: Any) -> int:
        """
        Use TinyTag to estimate the duration of the audio file and convert to tokens.

        Gemini estimates 32 tokens per second of audio
        https://ai.google.dev/gemini-api/docs/tokens?lang=python

View on GitHub (pinned to afd0fef371)

Solutions

1. Validate the source before constructing: len(audio_bytes) > 0 or os.path.getsize(path) > 0.
2. On URL-based blocks, verify the download (content-length, first bytes) before assignment and retry failures.
3. Repair the upstream fetch/upload that produced empty audio.

Example fix

# before
block = AudioBlock(path="clip.mp3")  # 0-byte file
data = block.resolve_audio(as_base64=True).read()

# after
if os.path.getsize("clip.mp3") == 0:
    raise RuntimeError("empty audio file")
block = AudioBlock(path="clip.mp3")

import os
ok = (audio_bytes and len(audio_bytes) > 0) or (path and os.path.getsize(path) > 0)
if not ok:
    raise ValueError("audio source is empty")

try:
    buf = block.resolve_audio(as_base64=True)
except ValueError as e:
    if "zero bytes" in str(e):
        refetch_or_skip()
    else:
        raise

When it happens

Trigger: AudioBlock(audio=b""), a 0-byte audio file path, or an audio URL returning an empty body; truncated uploads yielding empty files.

Common situations: Speech/audio ingestion pipelines where a fetch step wrote an empty file; expired presigned URLs returning empty 200s; test fixtures with placeholder empty audio.

Related errors

• resolve_image returned zero bytes
• resolve_video returned zero bytes
• resolve_document returned zero bytes
• LLM must be a FunctionCallingLLM
• No embeddings to aggregate

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/fcce399d2351f411. Report an issue: GitHub.