seed_vc.md

Seed-VC

Seed-VC is wired as --family seed_vc for voice conversion (vc) and singing voice conversion (svc). Every route takes a source audio file plus a target voice reference.

Common CLI shape:

audiocpp_cli --task vc --family seed_vc --model models/SeedVC-MLX --backend cuda --task-route <route> --audio source.wav --voice-ref target.wav --out converted.wav

Model

Field	Value
Family	seed_vc
Model directory	models/SeedVC-MLX
Tasks	vc, svc
Default vc route	v2_vc
Default svc route	v1_svc
Source audio	--audio
Target voice or singer	--voice-ref

V2 Voice Conversion

Use the V2 voice-conversion path for speech voice conversion. This route uses separate intelligibility and speaker-similarity CFG controls.

Field	Value
Task	vc
Route	v2_vc
Source audio	Required
Target voice reference	Required
Pitch controls	Not used
Style conversion	Not exposed in the current C++ route

audiocpp_cli --task vc --family seed_vc --model models/SeedVC-MLX --backend cuda --task-route v2_vc --audio source.wav --voice-ref target.wav --out converted.wav

Common Options (use directly)

Option	Values	Default	Meaning
--num-inference-steps	integer	30	V2 CFM denoising steps.
--seed	integer	random if omitted	Seed for V2 random sampling/noise.

Request Options (use with --request-option)

Option	Values	Default	Meaning
duration_scale	float > 0	1.0	Currently parsed and validated but does not affect V2 duration. Legacy length_adjust remains accepted.
intelligibility_guidance_scale	float	0.7	V2 classifier-free guidance scale for source-content intelligibility; default 0.7. Any finite value is accepted and used in V2 guidance arithmetic.
similarity_guidance_scale	float	0.7	V2 classifier-free guidance scale for target-speaker similarity; default 0.7. Any finite value is accepted and used in V2 guidance arithmetic.
voice_anonymization	bool	false	Use randomized voice conditioning instead of target-speaker conditioning.
noise_file	path	empty	Optional deterministic noise input for validation. Legacy noise_path remains accepted.

V1 Whisper + BigVGAN Voice Conversion

Use this V1 route when you want Whisper content features and the BigVGAN vocoder.

Field	Value
Task	vc
Route	v1_whisper_bigvgan_vc
Source audio	Required
Target voice reference	Required
Content path	Whisper
Vocoder	BigVGAN

audiocpp_cli --task vc --family seed_vc --model models/SeedVC-MLX --backend cuda --task-route v1_whisper_bigvgan_vc --audio source.wav --voice-ref target.wav --out converted.wav

Common Options (use directly)

Option	Values	Default	Meaning
--num-inference-steps	integer	30	V1 CFM denoising steps.
--seed	integer	random if omitted	Seed for V1 random noise.

Request Options (use with --request-option)

Option	Values	Default	Meaning
duration_scale	float > 0	1.0	Stretch or compress generated duration. Legacy length_adjust remains accepted.
inference_guidance_scale	float	0.7	V1 classifier-free guidance scale; default 0.7. Any finite value is accepted; V1 disables CFG for nonpositive values.

V1 XLSR + HiFT Voice Conversion

Use this V1 route when you want XLSR content features and the HiFT vocoder.

Field	Value
Task	vc
Route	v1_xlsr_hift_vc
Source audio	Required
Target voice reference	Required
Content path	XLSR
Vocoder	HiFT

audiocpp_cli --task vc --family seed_vc --model models/SeedVC-MLX --backend cuda --task-route v1_xlsr_hift_vc --audio source.wav --voice-ref target.wav --out converted.wav

Common Options (use directly)

Option	Values	Default	Meaning
--num-inference-steps	integer	30	V1 CFM denoising steps.
--seed	integer	random if omitted	Seed for V1 random noise.

Request Options (use with --request-option)

Option	Values	Default	Meaning
duration_scale	float > 0	1.0	Stretch or compress generated duration. Legacy length_adjust remains accepted.
inference_guidance_scale	float	0.7	V1 classifier-free guidance scale; default 0.7. Any finite value is accepted; V1 disables CFG for nonpositive values.

V1 Singing Voice Conversion

Use this route for singing voice conversion. It supports the V1 F0 and pitch controls.

Field	Value
Task	svc
Route	v1_svc
Source singing	Required
Target singer reference	Required
F0 extraction	Enabled by default; disable with f0_condition=false
Pitch controls	semitone_shift, auto_f0_adjust

audiocpp_cli --task svc --family seed_vc --model models/SeedVC-MLX --backend cuda --task-route v1_svc --audio singing.wav --voice-ref target.wav --out svc.wav

Common Options (use directly)

Option	Values	Default	Meaning
--num-inference-steps	integer	30	V1 CFM denoising steps.
--seed	integer	random if omitted	Seed for V1 random noise.

Request Options (use with --request-option)

Option	Values	Default	Meaning
duration_scale	float > 0	1.0	Stretch or compress generated duration. Legacy length_adjust remains accepted.
inference_guidance_scale	float	0.7	V1 classifier-free guidance scale; default 0.7. Any finite value is accepted; V1 disables CFG for nonpositive values.
f0_condition	bool	true	Enable F0-conditioned conversion.
auto_f0_adjust	bool	false	Automatically adjust F0 when F0 conditioning is enabled.
semitone_shift	integer	0	Shift pitch by semitones when F0 conditioning is enabled.

Common Options (use directly)

Option	Values	Default	Meaning
--task-route	v2_vc, v1_whisper_bigvgan_vc, v1_xlsr_hift_vc, v1_svc	v2_vc for vc; v1_svc for svc	Select the conversion route.
--audio	WAV path	required	Source speech or singing audio.
--voice-ref	WAV path	required	Target voice or singer reference.

Session Options (use with --session-option)

Option	Values	Default	Meaning
seed_vc.weight_type	native, f32, f16, bf16, q8_0	native	Weight storage type.

Compatibility

The schema-v1 option names are the preferred names. Existing local requests using the old names are still accepted through this compatibility map:

Old request option	New request option
inference_cfg_rate	inference_guidance_scale
intelligibility_cfg_rate	intelligibility_guidance_scale
similarity_cfg_rate	similarity_guidance_scale
anonymization_only	voice_anonymization
noise_path	noise_file
semi_tone_shift	semitone_shift