stt_service.proto

syntax = "proto3";

package sk.stt.v1;

import "google/protobuf/duration.proto";

service SttService {
  rpc StreamingRecognize (stream StreamingRecognitionRequest) returns (stream StreamingRecognitionResponse) {
  }
}

message StreamingRecognitionRequest {
  oneof streaming_request {
    RecognitionConfig config = 1;
    bytes audio_content = 2;
  }
}

message StreamingRecognitionResponse {
  repeated SpeechRecognitionChunk chunks = 1;
  reserved 2;
  reserved "end_of_single_utterance";
}

message RecognitionConfig {
  RecognitionSpec specification = 1;
}

message RecognitionSpec {
  enum AudioEncoding {
    AUDIO_ENCODING_UNSPECIFIED = 0;

    // 16-bit signed little-endian (Linear PCM)
    LINEAR16_PCM = 1;
  }

  AudioEncoding audio_encoding = 1;

  // 8000, 16000, 48000 only for pcm
  int64 sample_rate_hertz = 2;

  // code in BCP-47
  string language_code = 3;

  bool profanity_filter = 4;
  string model = 5;

  // If set true, tentative hypotheses may be returned as they become available (final=false flag)
  // If false or omitted, only final=true result(s) are returned.
  // Makes sense only for StreamingRecognize requests.
  bool partial_results = 7;

  bool single_utterance = 8;

  // This mark allows disable normalization text
  bool raw_results = 10;
}

message SpeechRecognitionChunk {
  repeated SpeechRecognitionAlternative alternatives = 1;
  // This flag shows that the received chunk contains a part of the recognized text that won't be changed.
  bool final = 2;
  // This flag shows that the received chunk is the end of an utterance.
  bool end_of_utterance = 3;
}

message SpeechRecognitionAlternative {
  string text = 1;
  float confidence = 2;
  repeated WordInfo words = 3;
}

message WordInfo {
  google.protobuf.Duration start_time = 1;
  google.protobuf.Duration end_time = 2;
  string word = 3;
  float confidence = 4;
}