RVC
Learn how to use the rvc_cli.py script to perform various operations with RVC.
Usage
To use the RVC CLI, navigate to the directory containing rvc_cli.py in your terminal and execute the script using the following syntax:
python rvc_cli.py <mode> [arguments]Replace <mode> with the desired mode of operation (e.g., infer, train, index) and provide the necessary arguments. For a detailed list of arguments available for each mode, run:
python rvc_cli.py <mode> -hThis will display a help message with explanations for each argument.
Modes
Infer
Performs a voice cloning conversion on a single audio file.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--pitch | Set the pitch of the audio. Higher values result in a higher pitch. | int | 0 | No |
--filter_radius | Apply median filtering to the extracted pitch values if this value is greater than or equal to three. This can help reduce breathiness in the output audio. | int | 3 | No |
--index_rate | Control the influence of the index file on the output. Higher values mean stronger influence. Lower values can help reduce artifacts but may result in less accurate voice cloning. | float | 0.3 | No |
--volume_envelope | Control the blending of the output's volume envelope. A value of 1 means the output envelope is fully used. | float | 1 | No |
--protect | Protect consonants and breathing sounds from artifacts. A value of 0.5 offers the strongest protection, while lower values may reduce the protection level but potentially mitigate the indexing effect. | float | 0.33 | No |
--hop_length | Only applicable for the Crepe pitch extraction method. Determines the time it takes for the system to react to a significant pitch change. Smaller values require more processing time but can lead to better pitch accuracy. | int | 128 | No |
--f0_method | Choose the pitch extraction algorithm for the conversion. 'rmvpe' is the default and generally recommended. | str | rmvpe | No |
--input_path | Full path to the input audio file. | str | Yes | |
--output_path | Full path to the output audio file. | str | Yes | |
--pth_path | Full path to the RVC model file (.pth). | str | Yes | |
--index_path | Full path to the index file (.index). | str | Yes | |
--split_audio | Split the audio into smaller segments before inference. This can improve the quality of the output for longer audio files. | bool | False | No |
--f0_autotune | Apply a light autotune to the inferred audio. Particularly useful for singing voice conversions. | bool | False | No |
--clean_audio | Clean the output audio using noise reduction algorithms. Recommended for speech conversions. | bool | False | No |
--clean_strength | Adjust the intensity of the audio cleaning process. Higher values result in stronger cleaning, but may lead to a more compressed sound. | float | 0.7 | No |
--export_format | Select the desired output audio format. | str | WAV | No |
--embedder_model | Choose the model used for generating speaker embeddings. | str | contentvec | No |
--embedder_model_custom | Specify the path to a custom model for speaker embedding. Only applicable if 'embedder_model' is set to 'custom'. | str | None | No |
--upscale_audio | Upscale the input audio to a higher quality before processing. This can improve the overall quality of the output, especially for low-quality input audio. | bool | False | No |
--f0_file | Full path to an external F0 file (.f0). This allows you to use pre-computed pitch values for the input audio. | str | None | No |
Batch Infer
Performs real-time voice cloning on all supported audio files within a specified folder. This mode utilizes the same arguments as the infer mode, except it requires an --input_folder and --output_folder instead of --input_path and --output_path, respectively.
| Argument | Description | Type | Required |
|---|---|---|---|
--input_folder | Path to the folder containing input audio files. | str | Yes |
--output_folder | Path to the folder for saving output audio files. | str | Yes |
TTS
Synthesizes text into speech using the specified voice and then applies voice conversion using the provided RVC model.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--tts_text | Text to be synthesized. | str | Yes | |
--tts_voice | Voice to be used for TTS synthesis. Refer to Microsoft's TTS voice list (opens in a new tab) for available options. | str | Yes | |
--tts_rate | Control the speaking rate of the TTS. Values range from -100 (slower) to 100 (faster). | int | 0 | No |
--output_tts_path | Full path to save the synthesized TTS audio. | str | Yes | |
--output_rvc_path | Full path to save the voice-converted audio using the synthesized TTS. | str | Yes |
This mode utilizes the same arguments as the infer mode for voice conversion settings.
Preprocess
Preprocesses a dataset for training an RVC model.
| Argument | Description | Type | Required |
|---|---|---|---|
--model_name | Name of the model to be trained. | str | Yes |
--dataset_path | Path to the dataset directory. | str | Yes |
--sample_rate | Target sampling rate for the audio data. | int | Yes |
--cpu_cores | Number of CPU cores to use for preprocessing. | int | No |
Extract
Extracts features from a dataset for training an RVC model.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--model_name | Name of the model. | str | Yes | |
--rvc_version | Version of the RVC model ('v1' or 'v2'). | str | v2 | No |
--f0_method | Pitch extraction method to use. | str | rmvpe | No |
--pitch_guidance | Enable or disable pitch guidance during feature extraction. | bool | True | No |
--hop_length | Hop length for feature extraction. Only applicable for Crepe pitch extraction. | int | 128 | No |
--cpu_cores | Number of CPU cores to use for feature extraction (optional). | int | None | No |
--sample_rate | Target sampling rate for the audio data. | int | Yes | |
--embedder_model | Choose the model used for generating speaker embeddings. | str | contentvec | No |
--embedder_model_custom | Specify the path to a custom model for speaker embedding. Only applicable if 'embedder_model' is set to 'custom'. | str | None | No |
Train
Trains an RVC model.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--model_name | Name of the model to be trained. | str | Yes | |
--rvc_version | Version of the RVC model to train ('v1' or 'v2'). | str | v2 | No |
--save_every_epoch | Save the model every specified number of epochs. | int | Yes | |
--save_only_latest | Save only the latest model checkpoint. | bool | False | No |
--save_every_weights | Save model weights every epoch. | bool | True | No |
--total_epoch | Total number of epochs to train for. | int | 1000 | No |
--sample_rate | Sampling rate of the training data. | int | Yes | |
--batch_size | Batch size for training. | int | 8 | No |
--gpu | GPU device to use for training (e.g., '0'). | str | 0 | No |
--pitch_guidance | Enable or disable pitch guidance during training. | bool | True | No |
--pretrained | Use a pretrained model for initialization. | bool | True | No |
--custom_pretrained | Use a custom pretrained model. | bool | False | No |
--g_pretrained_path | Path to the pretrained generator model file. | str | None | No |
--d_pretrained_path | Path to the pretrained discriminator model file. | str | None | No |
--overtraining_detector | Enable overtraining detection. | bool | False | No |
--overtraining_threshold | Threshold for overtraining detection. | int | 50 | No |
--sync_graph | Enable graph synchronization for distributed training. | bool | False | No |
--cache_data_in_gpu | Cache training data in GPU memory. | bool | False | No |
Index
Generates an index file for an RVC model.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--model_name | Name of the model. | str | Yes | |
--rvc_version | Version of the RVC model ('v1' or 'v2'). | str | v2 | No |
Model Extract
Extracts a checkpoint of the trained model.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--pth_path | Path to the main .pth model file. | str | Yes | |
--model_name | Name of the model. | str | Yes | |
--sample_rate | Sampling rate of the extracted model. | int | Yes | |
--pitch_guidance | Enable or disable pitch guidance for the extracted model. | bool | Yes | |
--rvc_version | Version of the extracted RVC model ('v1' or 'v2'). | str | v2 | No |
--epoch | Epoch number to extract from the model. | int | Yes | |
--step | Step number to extract from the model (optional). | int | None | No |
Model Information
Displays information about a trained model.
| Argument | Description | Type | Required |
|---|---|---|---|
--pth_path | Path to the .pth model file. | str | Yes |
Model Blender
Fuses two RVC models together.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--model_name | Name of the new fused model. | str | Yes | |
--pth_path_1 | Path to the first .pth model file. | str | Yes | |
--pth_path_2 | Path to the second .pth model file. | str | Yes | |
--ratio | Ratio for blending the two models (0.0 to 1.0). | float | 0.5 | No |
Tensorboard
Launches TensorBoard for monitoring training progress. This mode requires no arguments.
Download
Downloads a model from a provided link.
| Argument | Description | Type | Required |
|---|---|---|---|
--model_link | Direct link to the model file. | str | Yes |
Prerequisites
Installs prerequisites for RVC.
| Argument | Description | Type | Default | Required |
|---|---|---|---|---|
--pretraineds_v1 | Download pretrained models for RVC v1. | bool | True | No |
--pretraineds_v2 | Download pretrained models for RVC v2. | bool | True | No |
--models | Download additional models. | bool | True | No |
--exe | Download required executables. | bool | True | No |
Audio Analyzer
Analyzes an audio file and displays its information.
| Argument | Description | Type | Required |
|---|---|---|---|
--input_path | Path to the input audio file. | str | Yes |
Examples
Here are a few examples of how to use the RVC CLI:
- Inferring voice on an audio file:
python rvc_cli.py infer --pitch 5 --input_path "path/to/input.wav" --output_path "path/to/output.wav" --pth_path "path/to/model.pth" --index_path "path/to/index.index"- Training a new RVC model:
python rvc_cli.py train --model_name "my_model" --dataset_path "path/to/dataset" --sample_rate 48000 --total_epoch 500 --gpu 0- Generating an index file for a trained model:
python rvc_cli.py index --model_name "my_model"