Configuration Reference

Every parameter maps directly to client.process().

Cleanvoice resamples all audio to 44.1 kHz or 48 kHz before processing. Your original file is not modified — the cleaned output uses the resampled rate.

Option reference

Audio cleaning

Option	Type	Default	Description
`fillers`	`bool`	`False`	Remove "um", "uh", "like", and similar filler words
`long_silences`	`bool`	`False`	Trim long pauses and gaps between sentences
`mouth_sounds`	`bool`	`False`	Remove clicks, lip smacks, and tongue sounds
`breath`	`bool \| str`	`False`	Remove audible breathing between sentences
`stutters`	`bool`	`False`	Remove repeated word fragments ("I— I— I think")
`hesitations`	`bool`	`False`	Remove short hesitation sounds that aren't full filler words
`muted`	`bool`	`False`	Silence edits instead of cutting — preserves original timing

Filler and hesitation detection is language-aware. English, German, and Romanian have the most accurate models. See supported languages.

breath options:

Value	Behavior
`True`	Recommended for most audio. Best default for challenging recordings.
`"legacy"`	Conservative removal. Safer choice for already-clean recordings.
`"natural"`	Lighter touch — preserves more of the original breathing feel.
`False`	Disabled (default).

Recommended default: remove_noise=True and normalize=True. True, and omitting remove_noise, runs Noise v2. "legacy" pins the previous engine (DeepFilterNet). "v2" is still accepted and means the same as True. normalize=True is Normalize v2 (a boolean; there is no "v2" string). Filler, silence, and breath options stay off unless you ask for them.

result = client.process("episode.mp3", remove_noise=True, normalize=True)

Option	Type	Default	Description
`remove_noise`	`bool \| str`	`True` (Noise v2)	`True` is Noise v2 (recommended). Omitting it is the same as `True`. `"legacy"` pins the previous engine (DeepFilterNet). `"v2"` is still accepted. `False` turns it off.
`studio_sound`	`bool \| str`	`False`	Optional, more aggressive enhancement.
`normalize`	`bool`	`False`	Normalize v2 when `True`. Levels loudness. Off when omitted.
`keep_music`	`bool`	`False`	Preserve music sections during noise reduction
`autoeq`	`bool`	`False`	Legacy automatic EQ option. Prefer `studio_sound`; `autoeq` will be removed in a future release.

remove_noise values:

Value	Behavior
`True`	Recommended. Noise v2. This is also the default when the argument is omitted.
`"legacy"`	Previous engine (DeepFilterNet), pinned so it stays on that model.
`"v2"`	Still accepted. Same as `True` (Noise v2).
`False`	Off.

studio_sound options:

Value	Behavior
`True`	Aggressive studio-quality enhancement. Optional, and stronger than Remove Noise v2.
`"nightly"`	Advanced/experimental variant. Currently behaves similarly to `True`.
`False`	Disabled (default).

Output

Option	Type	Default	Description
`export_format`	`str`	`"auto"`	Audio-only output format: `mp3`, `wav`, `flac`, `m4a`, or `auto` (matches input). Video jobs keep the original video container format.
`target_lufs`	`float`	omitted (`-16` with Normalize v2)	Integrated loudness when `normalize=True`. Accepted range is `-36` to `-6`. `-16` is the podcast target and the value Normalize v2 uses when this argument is omitted.
`mute_lufs`	`float`	`-120`	Gate level for LUFS measurement. Default -120 disables gating.
`export_timestamps`	`bool`	`False`	Return a JSON file with edit markers for use in a DAW or NLE

To set a loudness target, pass target_lufs to the value you want:

result = client.process("episode.mp3", target_lufs=-16.0)

Content generation

Option	Type	Default	Description
`transcription`	`bool`	`False`	Full word-by-word transcript. Language auto-detected.
`summarize`	`bool`	`False`	Chapter markers, key learnings, and episode summary. Enables `transcription` automatically.
`social_content`	`bool`	`False`	Generate tweets, LinkedIn posts, and show notes. Enables `summarize` automatically.

Advanced

Option	Type	Default	Description
`signed_url`	`str`	`None`	Pre-signed PUT URL. Cleanvoice uploads directly to your storage instead of hosting the file.
`merge`	`bool`	`False`	Multi-track only. Merge all tracks into a single output file.
`audio_for_edl`	`bool`	`False`	Video workflows only. Return additional uncut enhanced audio alongside the edited video for EDL/NLE work.
`video`	`bool`	`False`	Process the input as video. The SDK auto-detects common video filenames and URLs, but explicit `video=True` is safest for ambiguous or extensionless URLs.

For raw REST requests, video must be set to true for actual video editing. Otherwise the file is treated as audio-only. In the Python SDK, common video paths are auto-detected, but being explicit is still safest.

SDK-only options

Option	Type	Default	Description
`output_path`	`str`	`None`	Save the cleaned audio to this path automatically
`progress_callback`	`callable`	`None`	Called with a dict payload during polling: `status`, `result`, `edit_id`, and `attempt`

def on_progress(update):
    print(f"Status: {update['status']}")

result = client.process(
    "episode.mp3",
    remove_noise=True,
    normalize=True,
    progress_callback=on_progress,
    output_path="cleaned.mp3",
)