syntax = "proto3";

package smartspeech.recognition.v1;

import "task.proto";

import "google/protobuf/duration.proto";

option go_package = "./;protocol";
option java_package = "TODO";

import "google/api/client.proto";
service SmartSpeech {
  option (google.api.default_host) = "sber.ru";
  rpc Recognize (stream RecognitionRequest) returns (stream RecognitionResponse);

  rpc AsyncRecognize (AsyncRecognizeRequest) returns (smartspeech.task.v1.Task);
}

message RecognitionRequest {
  oneof request {
    RecognitionOptions options = 1;
    bytes audio_chunk = 2;
  }
}

enum EouReason {
  UNSPECIFIED = 0; // Not an EOU
  ORGANIC = 1; // Proper EOU
  NO_SPEECH_TIMEOUT = 2;  // EOU because there was no speech for at least no_speech_timeout seconds
  MAX_SPEECH_TIMEOUT = 3;	// EOU because there was continuous speech with no breaks for at least max_speech_timeout seconds
}

message RecognitionResponse {
  repeated Hypothesis results = 1;
  bool eou = 2;  // marks final result for this utterance
  Emotions emotions_result = 3;  // may be set on end of utterance
  google.protobuf.Duration processed_audio_start = 4;  // starting position of processed audio
  google.protobuf.Duration processed_audio_end = 5;  // ending position of processed audio
  BackendInfo backend_info = 6;
  int32 channel = 7;
  SpeakerInfo speaker_info = 8;
  EouReason eou_reason = 9; // what caused the end of this utterance (eou must be true)
  string insight = 10; // json with insights response - will be filled in the last response
}

message AsyncRecognizeRequest {
  RecognitionOptions options = 1;
  string request_file_id = 2;
}

message RecognitionOptions {
  enum AudioEncoding {
    AUDIO_ENCODING_UNSPECIFIED = 0;
    PCM_S16LE = 1;  // 16-bit signed little-endian (Linear PCM)
    OPUS = 2;  // mime audio/ogg; codecs=opus
    MP3 = 3;  // MPEG-1/2 Layer-3
    FLAC = 4;
    ALAW = 5;
    MULAW = 6;
  }

  AudioEncoding audio_encoding = 1;
  int32 sample_rate = 2;  // For PCM_16LE, ALAW, MULAW audio encodings
  string language = 3;  // Language code in RFC-3066 format, i.e.: ru-RU
  string model = 4;
  int32 hypotheses_count = 5;
  Hints hints = 6;
  bool enable_profanity_filter = 7;
  bool enable_multi_utterance = 8;
  bool enable_partial_results = 9;
  google.protobuf.Duration no_speech_timeout = 10;
  google.protobuf.Duration max_speech_timeout = 11;
  int32 channels_count = 12;
  SpeakerSeparationOptions speaker_separation_options = 13;
  repeated string insight_models = 14;
  bool force_cyrillic = 15;
}

message Hints {
  repeated string words = 1;
  bool enable_letters = 2;
  google.protobuf.Duration eou_timeout = 3;
}

message SpeakerSeparationOptions {
  bool enable = 1;
  bool enable_only_main_speaker = 2;  // return only main speaker
  int32 count = 3;  // number of expected speakers
}

message Hypothesis {
  message WordAlignment {
    string word = 1;  // single word
    google.protobuf.Duration start = 2;  // starting position of the word in audio stream
    google.protobuf.Duration end = 3;  // ending position of the word in audio stream
  }

  string text = 1;  // non-normalized text result
  string normalized_text = 2;  // normalized text result
  google.protobuf.Duration start = 3;  // hypothesis starting position from current utterance start
  google.protobuf.Duration end = 4;  // hypothesis final position from current utterance start
  repeated WordAlignment word_alignments = 5;  // alignments of single words in audio stream
}

message Emotions {
  float positive = 1;  // confidence [0.0 - 1.0]
  float neutral = 2;
  float negative = 3;
}

message BackendInfo {
  string model_name = 1;
  string model_version = 2;
  string server_version = 3;
}

message SpeakerInfo {
  int32 speaker_id = 1;
  float main_speaker_confidence = 2;  // main speaker feature [0.0 - 1.0]
}