espnet2.fileio.omniio_audio.OmniIOAudioWriter

source

class espnet2.fileio.omniio_audio.OmniIOAudioWriter(outdir: Path | str, scpfile: Path | str, format: str = 'flac', bit_depth: int = 16, max_bin_size: int = 335544320)

Bases: object

Write audio into one Omni-IO archive and an scp pointing into it.

The archive is outdir/blob_<i>.bin + outdir/metadata.parquet with the columns that omniio.blob.blob.Blob writes (id, start_byte, end_byte, bin_index, path and the audio metadata of omniio.audio.write.audio_write), so other Omni-IO tools can read it. A new bin file is started once max_bin_size bytes would be exceeded. The writer owns outdir: archive files left there by an earlier run (blob_*.bin, metadata.parquet) are removed when it starts, so a rerun never leaves stale entries behind.

Examples

>>> with OmniIOAudioWriter('dump/format.1/wav', 'dump/format.1/wav.scp') as w:
...     w['utt1'] = 16000, numpy_array

• Parameters:
  • outdir – Archive directory.
  • scpfile – The scp file to write.
  • format – Audio codec inside the archive (“flac”, “wav” or “webm”).
  • bit_depth – PCM bit depth (ignored for “webm”).
  • max_bin_size – Maximum size of one bin file in bytes.

Start a new archive in outdir, removing one left by an earlier run.

close()

Finish the archive: write metadata.parquet and close the files.

An archive with no entry leaves no bin and no metadata file. Calling close() again does nothing.