RVC_CLI: Retrieval-based Voice Conversion Command Line Interface

Table of Contents

1. Installation
   • Windows
   • Linux
2. Getting Started
   • Inference
   • Training
   • UVR
   • Additional Features
3. API
4. Credits

Installation

Ensure that you have the necessary Python packages installed by following these steps (Python 3.9 is recommended):

Windows

Execute the install.bat file to activate a Conda environment. Afterward, launch the application using env/python.exe rvc.py instead of the conventional python rvc.py command.

Linux

chmod +x install.sh
./install.sh

Getting Started

Download the necessary models and executables by running the following command:

python rvc.py prerequisites

More information about the prerequisites command here

For detailed information and command-line options, refer to the help command:

python rvc.py -h

This command provides a clear overview of the available modes and their corresponding parameters, facilitating effective utilization of the RVC CLI.

Inference

Single Inference

python rvc.py infer --f0up_key "f0up_key" --filter_radius "filter_radius" --index_rate "index_rate" --hop_length "hop_length" --rms_mix_rate "rms_mix_rate" --protect "protect" --f0autotune "f0autotune" --f0method "f0method" --input_path "input_path" --output_path "output_path" --pth_path "pth_path" --index_path "index_path" --split_audio "split_audio" --clean_audio "clean_audio" --clean_strength "clean_strength" --export_format "export_format"

Parameter Name	Required	Default	Valid Options	Description
f0up_key	No	0	-24 to +24	Set the pitch of the audio, the higher the value, thehigher the pitch.
filter_radius	No	3	0 to 10	If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.
index_rate	No	0.3	0.0 to 1.0	Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.
hop_length	No	128	1 to 512	Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.
rms_mix_rate	No	1	0 to 1	Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.
protect	No	0.33	0 to 0.5	Safeguard distinct consonants and breathing sounds to prevent electro-acoustic tearing and other artifacts. Pulling the parameter to its maximum value of 0.5 offers comprehensive protection. However, reducing this value might decrease the extent of protection while potentially mitigating the indexing effect.
f0autotune	No	False	True or False	Apply a soft autotune to your inferences, recommended for singing conversions.
f0method	No	rmvpe	pm, harvest, dio, crepe, crepe-tiny, rmvpe, fcpe, hybrid[crepe+rmvpe], hybrid[crepe+fcpe], hybrid[rmvpe+fcpe], hybrid[crepe+rmvpe+fcpe]	Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.
input_path	Yes	None	Full path to the input audio file	Full path to the input audio file
output_path	Yes	None	Full path to the output audio file	Full path to the output audio file
pth_path	Yes	None	Full path to the pth file	Full path to the pth file
index_path	Yes	None	Full index file path	Full index file path
split_audio	No	False	True or False	Split the audio into chunks for inference to obtain better results in some cases.
clean_audio	No	False	True or False	Clean your audio output using noise detection algorithms, recommended for speaking audios.
clean_strength	No	0.7	0.0 to 1.0	Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.
export_format	No	WAV	WAV, MP3, FLAC, OGG, M4A	File audio format
embedder_model	No	hubert	hubert or contentvec	Embedder model to use for the audio conversion. The default model is hubert, which is recommended for most cases.
upscale_audio	No	False	True or False	Upscale the audio to 48kHz for better results.

Refer to python rvc.py infer -h for additional help.

Batch Inference

python rvc.py batch_infer --f0up_key "f0up_key" --filter_radius "filter_radius" --index_rate "index_rate" --hop_length "hop_length" --rms_mix_rate "rms_mix_rate" --protect "protect" --f0autotune "f0autotune" --f0method "f0method" --input_folder_path "input_folder_path" --output_folder_path "output_folder_path" --pth_path "pth_path" --index_path "index_path" --split_audio "split_audio" --clean_audio "clean_audio" --clean_strength "clean_strength" --export_format "export_format"

Parameter Name	Required	Default	Valid Options	Description
f0up_key	No	0	-24 to +24	Set the pitch of the audio, the higher the value, thehigher the pitch.
filter_radius	No	3	0 to 10	If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.
index_rate	No	0.3	0.0 to 1.0	Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.
hop_length	No	128	1 to 512	Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.
rms_mix_rate	No	1	0 to 1	Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.
protect	No	0.33	0 to 0.5	Safeguard distinct consonants and breathing sounds to prevent electro-acoustic tearing and other artifacts. Pulling the parameter to its maximum value of 0.5 offers comprehensive protection. However, reducing this value might decrease the extent of protection while potentially mitigating the indexing effect.
f0autotune	No	False	True or False	Apply a soft autotune to your inferences, recommended for singing conversions.
f0method	No	rmvpe	pm, harvest, dio, crepe, crepe-tiny, rmvpe, fcpe, hybrid[crepe+rmvpe], hybrid[crepe+fcpe], hybrid[rmvpe+fcpe], hybrid[crepe+rmvpe+fcpe]	Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.
input_folder_path	Yes	None	Full path to the input audio folder (The folder may only contain audio files)	Full path to the input audio folder
output_folder_path	Yes	None	Full path to the output audio folder	Full path to the output audio folder
pth_path	Yes	None	Full path to the pth file	Full path to the pth file
index_path	Yes	None	Full path to the index file	Full path to the index file
split_audio	No	False	True or False	Split the audio into chunks for inference to obtain better results in some cases.
clean_audio	No	False	True or False	Clean your audio output using noise detection algorithms, recommended for speaking audios.
clean_strength	No	0.7	0.0 to 1.0	Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.
export_format	No	WAV	WAV, MP3, FLAC, OGG, M4A	File audio format
embedder_model	No	hubert	hubert or contentvec	Embedder model to use for the audio conversion. The default model is hubert, which is recommended for most cases.
upscale_audio	No	False	True or False	Upscale the audio to 48kHz for better results.

Refer to python rvc.py batch_infer -h for additional help.

TTS Inference

python rvc.py tts_infer --tts_text "tts_text" --tts_voice "tts_voice" --f0up_key "f0up_key" --filter_radius "filter_radius" --index_rate "index_rate" --hop_length "hop_length" --rms_mix_rate "rms_mix_rate" --protect "protect" --f0autotune "f0autotune" --f0method "f0method" --output_tts_path "output_tts_path" --output_rvc_path "output_rvc_path" --pth_path "pth_path" --index_path "index_path"--split_audio "split_audio" --clean_audio "clean_audio" --clean_strength "clean_strength" --export_format "export_format"

Parameter Name	Required	Default	Valid Options	Description
tts_text	Yes	None	Text for TTS synthesis	Text for TTS synthesis
tts_voice	Yes	None	Voice for TTS synthesis	Voice for TTS synthesis
f0up_key	No	0	-24 to +24	Set the pitch of the audio, the higher the value, thehigher the pitch.
filter_radius	No	3	0 to 10	If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.
index_rate	No	0.3	0.0 to 1.0	Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.
hop_length	No	128	1 to 512	Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.
rms_mix_rate	No	1	0 to 1	Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.
protect	No	0.33	0 to 0.5	Safeguard distinct consonants and breathing sounds to prevent electro-acoustic tearing and other artifacts. Pulling the parameter to its maximum value of 0.5 offers comprehensive protection. However, reducing this value might decrease the extent of protection while potentially mitigating the indexing effect.
f0autotune	No	False	True or False	Apply a soft autotune to your inferences, recommended for singing conversions.
f0method	No	rmvpe	pm, harvest, dio, crepe, crepe-tiny, rmvpe, fcpe, hybrid[crepe+rmvpe], hybrid[crepe+fcpe], hybrid[rmvpe+fcpe], hybrid[crepe+rmvpe+fcpe]	Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.
output_tts_path	Yes	None	Full path to the output TTS audio file	Full path to the output TTS audio file
output_rvc_path	Yes	None	Full path to the input RVC audio file	Full path to the input RVC audio file
pth_path	Yes	None	Full path to the pth file	Full path to the pth file
index_path	Yes	None	Full path to the index file	Full path to the index file
split_audio	No	False	True or False	Split the audio into chunks for inference to obtain better results in some cases.
clean_audio	No	False	True or False	Clean your audio output using noise detection algorithms, recommended for speaking audios.
clean_strength	No	0.7	0.0 to 1.0	Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.
export_format	No	WAV	WAV, MP3, FLAC, OGG, M4A	File audio format
embedder_model	No	hubert	hubert or contentvec	Embedder model to use for the audio conversion. The default model is hubert, which is recommended for most cases.
upscale_audio	No	False	True or False	Upscale the audio to 48kHz for better results.

Refer to python rvc.py tts_infer -h for additional help.

Training

Preprocess Dataset

python rvc.py preprocess --model_name "model_name" --dataset_path "dataset_path" --sampling_rate "sampling_rate"

Parameter Name	Required	Default	Valid Options	Description
model_name	Yes	None	Name of the model	Name of the model
dataset_path	Yes	None	Full path to the dataset folder (The folder may only contain audio files)	Full path to the dataset folder
sampling_rate	Yes	None	32000, 40000, or 48000	Sampling rate of the audio data

Refer to python rvc.py preprocess -h for additional help.

Extract Features

python rvc.py extract --model_name "model_name" --rvc_version "rvc_version" --pitch_guidance "pitch_guidance" --hop_length "hop_length" --sampling_rate "sampling_rate"

Parameter Name	Required	Default	Valid Options	Description
model_name	Yes	None	Name of the model	Name of the model
rvc_version	No	v2	v1 or v2	Version of the model
pitch_guidance	No	True	True or False	By employing pitch guidance, it becomes feasible to mirror the intonation of the original voice, including its pitch. This feature is particularly valuable for singing and other scenarios where preserving the original melody or pitch pattern is essential.
hop_length	No	128	1 to 512	Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.
sampling_rate	Yes	None	32000, 40000, or 48000	Sampling rate of the audio data
embedder_model	No	hubert	hubert or contentvec	Embedder model to use for the audio conversion. The default model is hubert, which is recommended for most cases.

Start Training

python rvc.py train --model_name "model_name" --rvc_version "rvc_version" --save_every_epoch "save_every_epoch" --save_only_latest "save_only_latest" --save_every_weights "save_every_weights" --total_epoch "total_epoch" --sampling_rate "sampling_rate" --batch_size "batch_size" --gpu "gpu" --pitch_guidance "pitch_guidance" --overtraining_detector "overtraining_detector" --overtraining_threshold "overtraining_threshold"  --sync_graph "sync_graph" --pretrained "pretrained" --custom_pretrained "custom_pretrained" [--g_pretrained "g_pretrained"] [--d_pretrained "d_pretrained"]

Parameter Name	Required	Default	Valid Options	Description
model_name	Yes	None	Name of the model	Name of the model
rvc_version	No	v2	v1 or v2	Version of the model
save_every_epoch	Yes	None	1 to 50	Determine at how many epochs the model will saved at.
save_only_latest	No	False	True or False	Enabling this setting will result in the G and D files saving only their most recent versions, effectively conserving storage space.
save_every_weights	No	True	True or False	This setting enables you to save the weights of the model at the conclusion of each epoch.
total_epoch	No	1000	1 to 10000	Specifies the overall quantity of epochs for the model training process.
sampling_rate	Yes	None	32000, 40000, or 48000	Sampling rate of the audio data
batch_size	No	8	1 to 50	It's advisable to align it with the available VRAM of your GPU. A setting of 4 offers improved accuracy but slower processing, while 8 provides faster and standard results.
gpu	No	0	0 to ∞ separated by -	Specify the number of GPUs you wish to utilize for training by entering them separated by hyphens (-).
pitch_guidance	No	True	True or False	By employing pitch guidance, it becomes feasible to mirror the intonation of the original voice, including its pitch. This feature is particularly valuable for singing and other scenarios where preserving the original melody or pitch pattern is essential.
overtraining_detector	No	False	True or False	Utilize the overtraining detector to prevent overfitting. This feature is particularly valuable for scenarios where the model is at risk of overfitting.
overtraining_threshold	No	50	1 to 100	Set the threshold for the overtraining detector. The lower the value, the more sensitive the detector will be.
pretrained	No	True	True or False	Utilize pretrained models when training your own. This approach reduces training duration and enhances overall quality.
custom_pretrained	No	False	True or False	Utilizing custom pretrained models can lead to superior results, as selecting the most suitable pretrained models tailored to the specific use case can significantly enhance performance.
g_pretrained	No	None	Full path to pretrained file G, only if you have used custom_pretrained	Full path to pretrained file G
d_pretrained	No	None	Full path to pretrained file D, only if you have used custom_pretrained	Full path to pretrained file D
sync_graph	No	False	True or False	Synchronize the graph of the tensorbaord. Only enable this setting if you are training a new model.

Refer to python rvc.py train -h for additional help.

Generate Index File

python rvc.py index --model_name "model_name" --rvc_version "rvc_version"

Parameter Name	Required	Default	Valid Options	Description
model_name	Yes	None	Name of the model	Name of the model
rvc_version	Yes	None	v1 or v2	Version of the model

Refer to python rvc.py index -h for additional help.

UVR

python uvr.py [audio_file] [options]

Info and Debugging

Parameter Name	Required	Default	Valid Options	Description
audio_file	Yes	None	Any valid audio file path	The path to the audio file you want to separate, in any common format.
-d, --debug	No	False		Enable debug logging.
-e, --env_info	No	False		Print environment information and exit.
-l, --list_models	No	False		List all supported models and exit.
--log_level	No	info	info, debug, warning	Log level.

Separation I/O Params

Parameter Name	Required	Default	Valid Options	Description
-m, --model_filename	No	UVR-MDX-NET-Inst_HQ_3.onnx	Any valid model file path	Model to use for separation.
--output_format	No	WAV	Any common audio format	Output format for separated files.
--output_dir	No	None	Any valid directory path	Directory to write output files.
--model_file_dir	No	/tmp/audio-separator-models/	Any valid directory path	Model files directory.

Common Separation Parameters

Parameter Name	Required	Default	Valid Options	Description
--invert_spect	No	False		Invert secondary stem using spectrogram.
--normalization	No	0.9	Any float value	Max peak amplitude to normalize input and output audio to.
--single_stem	No	None	Instrumental, Vocals, Drums, Bass, Guitar, Piano, Other	Output only a single stem.
--sample_rate	No	44100	Any integer value	Modify the sample rate of the output audio.

MDXC Architecture Parameters

Parameter Name	Required	Default	Valid Options	Description
--mdxc_segment_size	No	256	Any integer value	Size of segments for MDXC architecture.
--mdxc_override_model_segment_size	No	False		Opverride model default segment size instead of using the model default value.
--mdxc_overlap	No	8	2 to 50	Amount of overlap between prediction windows for MDXC architecture.
--mdxc_batch_size	No	1	Any integer value	Batch size for MDXC architecture.
--mdxc_pitch_shift	No	0	Any integer value	Shift audio pitch by a number of semitones while processing for MDXC architecture.

MDX Architecture Parameters

Parameter Name	Required	Default	Valid Options	Description
--mdx_segment_size	No	256	Any integer value	Size of segments for MDX architecture.
--mdx_overlap	No	0.25	0.001 to 0.999	Amount of overlap between prediction windows for MDX architecture.
--mdx_batch_size	No	1	Any integer value	Batch size for MDX architecture.
--mdx_hop_length	No	1024	Any integer value	Hop length for MDX architecture.
--mdx_enable_denoise	No	False		Enable denoising during separation for MDX architecture.

Demucs Architecture Parameters

Parameter Name	Required	Default	Valid Options	Description
--demucs_segment_size	No	Default	Any integer value	Size of segments for Demucs architecture.
--demucs_shifts	No	2	Any integer value	Number of predictions with random shifts for Demucs architecture.
--demucs_overlap	No	0.25	0.001 to 0.999	Overlap between prediction windows for Demucs architecture.
--demucs_segments_enabled	No	True		Enable segment-wise processing for Demucs architecture.

VR Architecture Parameters

Parameter Name	Required	Default	Valid Options	Description
--vr_batch_size	No	4	Any integer value	Batch size for VR architecture.
--vr_window_size	No	512	Any integer value	Window size for VR architecture.
--vr_aggression	No	5	-100 to 100	Intensity of primary stem extraction for VR architecture.
--vr_enable_tta	No	False		Enable Test-Time-Augmentation for VR architecture.
--vr_high_end_process	No	False		Mirror the missing frequency range of the output for VR architecture.
--vr_enable_post_process	No	False		Identify leftover artifacts within vocal output for VR architecture.
--vr_post_process_threshold	No	0.2	0.1 to 0.3	Threshold for post-process feature for VR architecture.

Additional Features

Model Extract

python rvc.py model_extract --pth_path "pth_path" --model_name "model_name" --sampling_rate "sampling_rate" --pitch_guidance "pitch_guidance" --rvc_version "rvc_version" --epoch "epoch" --step "step"

Parameter Name	Required	Default	Valid Options	Description
pth_path	Yes	None	Path to the pth file	Full path to the pth file
model_name	Yes	None	Name of the model	Name of the model
sampling_rate	Yes	None	32000, 40000, or 48000	Sampling rate of the audio data
pitch_guidance	Yes	None	True or False	By employing pitch guidance, it becomes feasible to mirror the intonation of the original voice, including its pitch. This feature is particularly valuable for singing and other scenarios where preserving the original melody or pitch pattern is essential.
rvc_version	Yes	None	v1 or v2	Version of the model
epoch	Yes	None	1 to 10000	Specifies the overall quantity of epochs for the model training process.
step	Yes	None	1 to ∞	Specifies the overall quantity of steps for the model training process.

Model Information

python rvc.py model_information --pth_path "pth_path"

Parameter Name	Required	Default	Valid Options	Description
pth_path	Yes	None	Path to the pth file	Full path to the pth file

Model Blender

python rvc.py model_blender --model_name "model_name" --pth_path_1 "pth_path_1" --pth_path_2 "pth_path_2" --ratio "ratio"

Parameter Name	Required	Default	Valid Options	Description
model_name	Yes	None	Name of the model	Name of the model
pth_path_1	Yes	None	Path to the first pth file	Full path to the first pth file
pth_path_2	Yes	None	Path to the second pth file	Full path to the second pth file
ratio	No	0.5	0.0 to 1	Value for blender ratio

Launch TensorBoard

python rvc.py tensorboard

Download Models

Run the download script with the following command:

python rvc.py download --model_link "model_link"

Parameter Name	Required	Default	Valid Options	Description
model_link	Yes	None	Link of the model (enclosed in double quotes; Google Drive or Hugging Face)	Link of the model

Refer to python rvc.py download -h for additional help.

Audio Analyzer

python rvc.py audio_analyzer --input_path "input_path"

Parameter Name	Required	Default	Valid Options	Description
input_path	Yes	None	Full path to the input audio file	Full path to the input audio file

Refer to python rvc.py audio_analyzer -h for additional help.

Prerequisites Download

python rvc.py prerequisites --pretraineds_v1 "pretraineds_v1" --pretraineds_v2 "--pretraineds_v2" --models "models" --exe "exe"

Parameter Name	Required	Default	Valid Options	Description
pretraineds_v1	No	True	True or False	Download pretrained models for v1
pretraineds_v2	No	True	True or False	Download pretrained models for v2
models	No	True	True or False	Download models for v1 and v2
exe	No	True	True or False	Download the necessary executable files for the CLI to function properly (FFmpeg and FFprobe)

API

python rvc.py api --host "host" --port "port"

Parameter Name	Required	Default	Valid Options	Description
host	No	127.0.0.1	Value for host IP	Value for host IP
port	No	8000	Value for port number	Value for port number

To use the RVC CLI via the API, utilize the provided script. Make API requests to the following endpoints:

• Docs: /docs
• Ping: /ping
• Infer: /infer
• Batch Infer: /batch_infer
• TTS: /tts
• Preprocess: /preprocess
• Extract: /extract
• Train: /train
• Index: /index
• Model Information: /model_information
• Model Fusion: /model_fusion
• Download: /download

Make POST requests to these endpoints with the same required parameters as in CLI mode.

Credits

The RVC CLI builds upon the foundations of the following projects:

• ContentVec by auspicious3000
• HIFIGAN by jik876
• audio-slicer by openvpi
• python-audio-separator by karaokenerds
• RMVPE by Dream-High
• FCPE by CNChTu
• VITS by jaywalnut310
• So-Vits-SVC by svc-develop-team
• Harmonify by Eempostor
• Retrieval-based-Voice-Conversion-WebUI by RVC-Project
• Mangio-RVC-Fork by Mangio621
• anyf0 by SoulMelody

We acknowledge and appreciate the contributions of the respective authors and communities involved in these projects.