Manifest¶

Manifests are JSONL files. Each line is one JSON object.

Conversion Manifest¶

build_manifest combines conversion results and validation results.

Fields:

Field	Type	Meaning
`source_path`	string	Original source file path.
`output_path`	string or null	Converted output path, when available.
`status`	string	`ok`, `conversion_failed`, or `validation_failed`.
`duration_sec`	number or null	Validated duration.
`sample_rate`	number or null	Validated sample rate.
`channels`	number or null	Validated channel count.
`error`	string or null	Error or validation reason.

Example:

{"source_path":"data/raw_mp3/a.mp3","output_path":"data/wav16k/a.wav","status":"ok","duration_sec":1.0,"sample_rate":16000,"channels":1,"error":null}

Chunk Manifest¶

build_chunk_manifest records one row per source file.

Fields:

Field	Type	Meaning
`source_path`	string	Original source file path.
`status`	string	`ok` or `chunking_failed`.
`num_chunks`	number	Number of chunks written.
`chunk_paths`	list[string]	Written chunk paths.
`error`	string or null	Chunking error, when any.

Writing Manifests¶

CLI:

audio-prep convert \
    --input-dir data/raw_mp3 \
    --output-dir data/wav16k \
    --manifest data/manifest.jsonl

audio-prep chunk \
    --input-dir data/raw_mp3 \
    --output-dir data/chunks \
    --manifest data/chunk_manifest.jsonl

Python:

Use write_manifest for both conversion and chunk records:

from audio_prep.manifest import write_manifest

write_manifest(records, "data/manifest.jsonl")

Parent directories are created automatically.