FLAC decoder loses last frame data when final frame is incomplete/corrupted

[autumn-2-net]

FLAC decoder loses last frame data when final frame is incomplete/corrupted

Summary

When decoding FLAC files with incomplete or slightly corrupted final frames, PyAV's decoder silently drops the last frame's data, even when error detection is disabled via err_detect='ignore_err'. This behavior differs from FFmpeg CLI and other audio libraries (e.g., torchaudio), which successfully decode the complete audio including the last frame.

Environment

• PyAV version: 16.0.1
• FFmpeg version: 2023-09-29-git-40aa451154-full
• Python version: 3.10
• OS: Windows

Expected Behavior

PyAV should decode all available audio data from the FLAC file, including partial data from the last frame, similar to how FFmpeg CLI handles it with -err_detect ignore_err.

Actual Behavior

PyAV raises av.error.InvalidDataError on the last frame and discards all data from that frame, resulting in missing audio samples (typically 80-100ms of audio).

Comparison with Other Tools

Tool	Behavior	Samples Decoded
torchaudio	✅ Decodes completely	13,561,718
FFmpeg CLI	✅ Decodes completely	13,561,718
PyAV	❌ Loses last frame	13,557,760
Missing	-	3,958 samples (~82ms)

Minimal Reproducible Example

import av
import torchaudio

# Test file (replace with your FLAC file that has incomplete last frame)
input_file = 'test.flac'

# Reference: Load with torchaudio (complete decoding)
waveform_ref, sr_ref = torchaudio.load(input_file)
reference_samples = waveform_ref.shape[1]
print(f"torchaudio samples: {reference_samples:,}")

# PyAV: Attempt to decode with error ignoring
options = {
    'err_detect': 'ignore_err',
    'fflags': 'genpts'
}

container = av.open(input_file, options=options)
stream = container.streams.audio[0]

# Also try setting options on codec context
stream.codec_context.options = {'err_detect': 'ignore_err'}

total_samples = 0
frame_count = 0
errors = 0

for packet in container.demux(stream):
    try:
        for frame in packet.decode():
            frame_count += 1
            # total_samples += frame.samples  # or frame.duration in newer versions
            total_samples += frame.duration  # or frame.duration in newer versions
    except av.error.InvalidDataError as e:
        errors += 1
        print(f"InvalidDataError on frame {frame_count}: {e}")
        # Even with error handling, the last frame's data is lost
        continue

container.close()

print(f"PyAV samples: {total_samples:,}")
print(f"Missing samples: {reference_samples - total_samples:,}")
print(f"Missing duration: {(reference_samples - total_samples) / sr_ref * 1000:.2f} ms")
print(f"Errors encountered: {errors}")

output:

torchaudio samples: 13,561,718
InvalidDataError on frame 3310: [Errno 1094995529] Invalid data found when processing input: 'avcodec_receive_frame()'
PyAV samples: 13,557,760
Missing samples: 3,958
Missing duration: 82.46 ms
Errors encountered: 1