realtime-voice-conversion/protos/audio.proto

syntax = "proto3";
package voder_audio;

message VocodeAudioRequest {
  repeated float float_audio = 1;

  // An autoincrement sent with each request.
  int64 request_batch_number = 2;

  message VocodeParams {
    // The initial sample rate coming from the microphone.
    int32 initial_sample_rate = 1;

    // Original sample
    int32 original_source_rate = 9;
    bool original_source_save_file = 10;

    // Resample before passing to the algorithm?
    bool pre_convert_resample = 2;
    int32 pre_convert_resample_rate = 3;
    bool pre_convert_resample_save_file = 4;

    // The hard-coded hyperparameter of the model.
    // Probably don't want to change from 16000.
    int32 model_hyperparameter_sampling_rate = 5;
    bool model_save_file = 11;

    // Resample after passing to the algorithm?
    bool post_convert_resample = 6;
    int32 post_convert_resample_rate = 7;
    bool post_convert_resample_save_file = 8;
  }

  VocodeParams vocode_params = 10;

  // We'll resend these params for every batch, but will only use
  // the last batch's settings. Kind of a dumb API, but it'll work.
  //int32 sample_rate = 2;

  //bool skip_resample = 3;
  bool skip_vocode = 4;

  // Debugging
  //bool save_files = 5;

  // How big we let the buffer grow before running 'convert'.
  int32 buffer_size_minimum = 6;

  // Requested output rate
  //int32 output_rate = 7;

  // Discard the vocoded audio and return the original
  bool discard_vocoded_audio = 8;

  // Rate param for the model
  //int32 model_sampling_rate = 9;
}

message VocodeAudioResponse {
  repeated float float_audio = 1;

  // A sidecar-side autoincrement sent with each request.
  int64 response_batch_number = 2;
}