[BUG] zstandard library is missing. CLAP analysis doesn't use fallback WAV files.

[sorcery88]

I have some m4a files that are failing to read, and fallback processing is failing.

Here's an example as installed,

audiomuse-ai-worker  | [INFO]-[11-03-2026 22-19-26]-Downloaded 'Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame;' to '/app/temp_audio/gLwaaiNqQGKTJRtXpjGmOa.m4a'
audiomuse-ai-worker  | [INFO]-[11-03-2026 22-19-26]-Starting analysis for: gLwaaiNqQGKTJRtXpjGmOa.m4a
audiomuse-ai-worker  | /app/tasks/analysis.py:212: UserWarning: PySoundFile failed. Trying audioread instead.
audiomuse-ai-worker  |   audio, sr = librosa.load(file_path, sr=target_sr, mono=True, duration=AUDIO_LOAD_TIMEOUT)
audiomuse-ai-worker  | /usr/local/lib/python3.12/dist-packages/librosa/core/audio.py:184: FutureWarning: librosa.core.audio.__audioread_load
audiomuse-ai-worker  |  Deprecated as of librosa version 0.10.0.
audiomuse-ai-worker  |  It will be removed in librosa version 1.0.
audiomuse-ai-worker  |   y, sr_native = __audioread_load(path, offset, duration, dtype)
audiomuse-ai-worker  | [WARNING]-[11-03-2026 22-19-26]-Direct librosa load failed for gLwaaiNqQGKTJRtXpjGmOa.m4a: float division by zero. Attempting fallback conversion.
audiomuse-ai-worker  | [ERROR]-[11-03-2026 22-19-26]-Fallback loading method also failed for gLwaaiNqQGKTJRtXpjGmOa.m4a: `zstandard` is required to use zstd compression
audiomuse-ai-worker  | [WARNING]-[11-03-2026 22-19-26]-Could not load a valid audio signal for gLwaaiNqQGKTJRtXpjGmOa.m4a after all attempts. Skipping track.
audiomuse-ai-worker  | [WARNING]-[11-03-2026 22-19-26]-Skipping track Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame; by Dedicated Men Of Zion as analysis returned None.

I tried "pip install zstandard" to overcome the most obvious issue. Processing gets further - we get MusicCNN analysis. CLAP doesn't seem to use the generated WAV file from earlier though so still fails.

audiomuse-ai-worker  | [INFO]-[12-03-2026 00-14-49]-Downloaded 'Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame;' to '/app/temp_audio/gLwaaiNqQGKTJRtXpjGmOa.m4a'
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-14-50]-Starting analysis for: gLwaaiNqQGKTJRtXpjGmOa.m4a
audiomuse-ai-worker  | /app/tasks/analysis.py:212: UserWarning: PySoundFile failed. Trying audioread instead.
audiomuse-ai-worker  |   audio, sr = librosa.load(file_path, sr=target_sr, mono=True, duration=AUDIO_LOAD_TIMEOUT)
audiomuse-ai-worker  | /usr/local/lib/python3.12/dist-packages/librosa/core/audio.py:184: FutureWarning: librosa.core.audio.__audioread_load
audiomuse-ai-worker  |  Deprecated as of librosa version 0.10.0.
audiomuse-ai-worker  |  It will be removed in librosa version 1.0.
audiomuse-ai-worker  |   y, sr_native = __audioread_load(path, offset, duration, dtype)
audiomuse-ai-worker  | [WARNING]-[12-03-2026 00-14-50]-Direct librosa load failed for gLwaaiNqQGKTJRtXpjGmOa.m4a: float division by zero. Attempting fallback conversion.
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-14-55]-Fallback: Pre-processing gLwaaiNqQGKTJRtXpjGmOa.m4a to a smaller WAV for safe loading...
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-14-57]-Fallback: Converted gLwaaiNqQGKTJRtXpjGmOa.m4a to temporary WAV for robust loading.
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-14-59]-CUDA provider not available - using CPU only
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-SUCCESSFULLY ANALYZED 'Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame; by Dedicated Men Of Zion' (ID: gLwaaiNqQGKTJRtXpjGmOa):
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-  - Tempo: 125.00, Energy: 0.1007, Key: D minor
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-  - Top Moods: {'rock': 0.5587335228919983, 'blues': 0.5374235510826111, 'soul': 0.5321893692016602, 'Hip-Hop': 0.5240907669067383, 'classic rock': 0.5204477906227112}
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-  - Other Features: danceable:0.35,aggressive:0.11,happy:0.36,party:0.32,relaxed:0.73,sad:0.74
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-  - Starting CLAP analysis for Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame; by Dedicated Men Of Zion...
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-Lazy-loading CLAP audio model on first use...
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-Loading CLAP audio model from /app/model/clap_audio_model.onnx...
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-CLAP Audio: Using ONNX Runtime automatic thread management
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-07]-CUDA provider not available - using CPU only
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-08]-✓ CLAP audio model loaded successfully (~268MB)
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-08]-✓ CLAP audio model initialized successfully (for music analysis)
audiomuse-ai-worker  | /app/tasks/clap_analyzer.py:578: UserWarning: PySoundFile failed. Trying audioread instead.
audiomuse-ai-worker  |   audio_data, sr = librosa.load(audio_path, sr=SAMPLE_RATE, mono=True, duration=AUDIO_LOAD_TIMEOUT)
audiomuse-ai-worker  | /usr/local/lib/python3.12/dist-packages/librosa/core/audio.py:184: FutureWarning: librosa.core.audio.__audioread_load
audiomuse-ai-worker  |  Deprecated as of librosa version 0.10.0.
audiomuse-ai-worker  |  It will be removed in librosa version 1.0.
audiomuse-ai-worker  |   y, sr_native = __audioread_load(path, offset, duration, dtype)
audiomuse-ai-worker  | [ERROR]-[12-03-2026 00-15-08]-Failed to load audio for CLAP analysis (/app/temp_audio/gLwaaiNqQGKTJRtXpjGmOa.m4a) with AUDIO_LOAD_TIMEOUT=600: float division by zero
audiomuse-ai-worker  | Traceback (most recent call last):
audiomuse-ai-worker  |   File "/app/tasks/clap_analyzer.py", line 578, in analyze_audio_file
audiomuse-ai-worker  |     audio_data, sr = librosa.load(audio_path, sr=SAMPLE_RATE, mono=True, duration=AUDIO_LOAD_TIMEOUT)
audiomuse-ai-worker  |                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
audiomuse-ai-worker  |   File "/usr/local/lib/python3.12/dist-packages/librosa/core/audio.py", line 193, in load
audiomuse-ai-worker  |     y = resample(y, orig_sr=sr_native, target_sr=sr, res_type=res_type)
audiomuse-ai-worker  |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
audiomuse-ai-worker  |   File "/usr/local/lib/python3.12/dist-packages/librosa/core/audio.py", line 633, in resample
audiomuse-ai-worker  |     ratio = float(target_sr) / orig_sr
audiomuse-ai-worker  |             ~~~~~~~~~~~~~~~~~^~~~~~~~~
audiomuse-ai-worker  | ZeroDivisionError: float division by zero
audiomuse-ai-worker  | [INFO]-[12-03-2026 00-15-09]-✓ CLAP model(s) unloaded from memory (~268MB freed + GPU memory released)

Are my files very weird? Here's what ffmpeg -i says for the same file above:

ffmpeg version 4.1.8 Copyright (c) 2000-2021 the FFmpeg developers
  built with gcc 8.5.0 (GCC)
  configuration: --prefix=/usr --incdir='${prefix}/include/ffmpeg' --arch=i686 --target-os=linux --cross-prefix=/usr/local/i686-pc-linux-gnu/bin/i686-pc-linux-gnu- --enable-cross-compile --enable-optimizations --enable-pic --enable-gpl --enable-shared --disable-static --disable-stripping --enable-version3 --enable-encoders --enable-pthreads --disable-protocols --disable-protocol=rtp --enable-protocol=file --enable-protocol=pipe --disable-muxer=image2 --disable-muxer=image2pipe --disable-swscale-alpha --disable-ffplay --disable-ffprobe --disable-doc --disable-devices --disable-bzlib --disable-altivec --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libmp3lame --disable-vaapi --disable-cuvid --disable-nvenc --disable-decoder=aac --disable-decoder=aac_fixed --disable-encoder=aac --disable-decoder=amrnb --disable-decoder=ac3 --disable-decoder=ac3_fixed --disable-encoder=zmbv --disable-encoder=dca --disable-decoder=dca --disable-encoder=ac3 --disable-encoder=ac3_fixed --disable-encoder=eac3 --disable-decoder=eac3 --disable-encoder=truehd --disable-decoder=truehd --disable-encoder=hevc_vaapi --disable-decoder=hevc --disable-muxer=hevc --disable-demuxer=hevc --disable-parser=hevc --disable-bsf=hevc_mp4toannexb --x86asmexe=yasm --cc=/usr/local/i686-pc-linux-gnu/bin/i686-pc-linux-gnu-wrap-gcc --enable-yasm --enable-libx264 --enable-encoder=libx264
  libavutil      56. 22.100 / 56. 22.100
  libavcodec     58. 35.100 / 58. 35.100
  libavformat    58. 20.100 / 58. 20.100
  libavdevice    58.  5.100 / 58.  5.100
  libavfilter     7. 40.101 /  7. 40.101
  libswscale      5.  3.100 /  5.  3.100
  libswresample   3.  3.100 /  3.  3.100
  libpostproc    55.  3.100 / 55.  3.100
Guessed Channel Layout for Input Stream #0.0 : stereo
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from '20210116-dedicated-men-of-zion-tiny-desk-home-concert.m4a':
  Metadata:
    major_brand     : M4A
    minor_version   : 512
    compatible_brands: isomiso2
    title           : Father, Guide Me, Teach Me; Can't Turn Me Around; It's A Shame;
    artist          : Dedicated Men Of Zion
    album_artist    : Various Artists
    album           : Tiny Desk Concerts 2021
    encoder         : Lavf58.20.100
    track           : 6
  Duration: 00:13:00.54, start: 0.000000, bitrate: 113 kb/s
    Stream #0:0(eng): Audio: aac (mp4a / 0x6134706D), 44100 Hz, stereo, 112 kb/s (default)
    Metadata:
      handler_name    : SoundHandler

Expected behavior
zstandard is installed so fallback processing works
CLAP analysis uses the fallback .wav so it can also succeed

Environment

• Docker image - ghcr.io/neptunehub/audiomuse-ai:latest
• Ubuntu based docker compose install, WSL worker container also experiencing the same issue

[NeptuneHub]

Hi and thanks for the feedback. About the two different issue that you shared:

• Missing zstandard library => I just added;
• CLAP analysis that didn't use the fallback method, I just fixed.

The fix is now in devel branch and the new image should be ready in 30-60 minute.

Important: in this devel version I'm releasing a different machine learning model in place of clap that need a fresh installation an a new rescan. So I suggest for now to use it in a different deployment. If instead you're a new user, instead, is good that you start the analysis directly with the last model.

[NeptuneHub]

Released in version 0.9.0 also.