embedded/v1alpha1/embedded_assistant.proto

// Copyright 2017 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

syntax = "proto3";

package google.assistant.embedded.v1alpha1;

import "google/api/annotations.proto";
import "google/rpc/status.proto";

option go_package = "google.golang.org/genproto/googleapis/assistant/embedded/v1alpha1;embedded";
option java_multiple_files = true;
option java_outer_classname = "AssistantProto";
option java_package = "com.google.assistant.embedded.v1alpha1";

// Service that implements Google Assistant API.
service EmbeddedAssistant {
  // Initiates or continues a conversation with the embedded assistant service.
  // Each call performs one round-trip, sending an audio request to the service
  // and receiving the audio response. Uses bidirectional streaming to receive
  // results, such as the `END_OF_UTTERANCE` event, while sending audio.
  //
  // A conversation is one or more gRPC connections, each consisting of several
  // streamed requests and responses.
  // For example, the user says *Add to my shopping list* and the assistant
  // responds *What do you want to add?*. The sequence of streamed requests and
  // responses in the first gRPC message could be:
  //
  // *   ConverseRequest.config
  // *   ConverseRequest.audio_in
  // *   ConverseRequest.audio_in
  // *   ConverseRequest.audio_in
  // *   ConverseRequest.audio_in
  // *   ConverseResponse.event_type.END_OF_UTTERANCE
  // *   ConverseResponse.result.microphone_mode.DIALOG_FOLLOW_ON
  // *   ConverseResponse.audio_out
  // *   ConverseResponse.audio_out
  // *   ConverseResponse.audio_out
  //
  // The user then says *bagels* and the assistant responds
  // *OK, I've added bagels to your shopping list*. This is sent as another gRPC
  // connection call to the `Converse` method, again with streamed requests and
  // responses, such as:
  //
  // *   ConverseRequest.config
  // *   ConverseRequest.audio_in
  // *   ConverseRequest.audio_in
  // *   ConverseRequest.audio_in
  // *   ConverseResponse.event_type.END_OF_UTTERANCE
  // *   ConverseResponse.result.microphone_mode.CLOSE_MICROPHONE
  // *   ConverseResponse.audio_out
  // *   ConverseResponse.audio_out
  // *   ConverseResponse.audio_out
  // *   ConverseResponse.audio_out
  //
  // Although the precise order of responses is not guaranteed, sequential
  // ConverseResponse.audio_out messages will always contain sequential portions
  // of audio.
  rpc Converse(stream ConverseRequest) returns (stream ConverseResponse);
}

// Specifies how to process the `ConverseRequest` messages.
message ConverseConfig {
  // *Required* Specifies how to process the subsequent incoming audio.
  AudioInConfig audio_in_config = 1;

  // *Required* Specifies how to format the audio that will be returned.
  AudioOutConfig audio_out_config = 2;

  // *Required* Represents the current dialog state.
  ConverseState converse_state = 3;
}

// Specifies how to process the `audio_in` data that will be provided in
// subsequent requests. For recommended settings, see the Google Assistant SDK
// [best
// practices](https://developers.google.com/assistant/sdk/develop/grpc/best-practices/audio).
message AudioInConfig {
  // Audio encoding of the data sent in the audio message.
  // Audio must be one-channel (mono). The only language supported is "en-US".
  enum Encoding {
    // Not specified. Will return result [google.rpc.Code.INVALID_ARGUMENT][].
    ENCODING_UNSPECIFIED = 0;

    // Uncompressed 16-bit signed little-endian samples (Linear PCM).
    // This encoding includes no header, only the raw audio bytes.
    LINEAR16 = 1;

    // [`FLAC`](https://xiph.org/flac/documentation.html) (Free Lossless Audio
    // Codec) is the recommended encoding because it is
    // lossless--therefore recognition is not compromised--and
    // requires only about half the bandwidth of `LINEAR16`. This encoding
    // includes the `FLAC` stream header followed by audio data. It supports
    // 16-bit and 24-bit samples, however, not all fields in `STREAMINFO` are
    // supported.
    FLAC = 2;
  }

  // *Required* Encoding of audio data sent in all `audio_in` messages.
  Encoding encoding = 1;

  // *Required* Sample rate (in Hertz) of the audio data sent in all `audio_in`
  // messages. Valid values are from 16000-24000, but 16000 is optimal.
  // For best results, set the sampling rate of the audio source to 16000 Hz.
  // If that's not possible, use the native sample rate of the audio source
  // (instead of re-sampling).
  int32 sample_rate_hertz = 2;
}

// Specifies the desired format for the server to use when it returns
// `audio_out` messages.
message AudioOutConfig {
  // Audio encoding of the data returned in the audio message. All encodings are
  // raw audio bytes with no header, except as indicated below.
  enum Encoding {
    // Not specified. Will return result [google.rpc.Code.INVALID_ARGUMENT][].
    ENCODING_UNSPECIFIED = 0;

    // Uncompressed 16-bit signed little-endian samples (Linear PCM).
    LINEAR16 = 1;

    // MP3 audio encoding. The sample rate is encoded in the payload.
    MP3 = 2;

    // Opus-encoded audio wrapped in an ogg container. The result will be a
    // file which can be played natively on Android and in some browsers (such
    // as Chrome). The quality of the encoding is considerably higher than MP3
    // while using the same bitrate. The sample rate is encoded in the payload.
    OPUS_IN_OGG = 3;
  }

  // *Required* The encoding of audio data to be returned in all `audio_out`
  // messages.
  Encoding encoding = 1;

  // *Required* The sample rate in Hertz of the audio data returned in
  // `audio_out` messages. Valid values are: 16000-24000.
  int32 sample_rate_hertz = 2;

  // *Required* Current volume setting of the device's audio output.
  // Valid values are 1 to 100 (corresponding to 1% to 100%).
  int32 volume_percentage = 3;
}

// Provides information about the current dialog state.
message ConverseState {
  // *Required* The `conversation_state` value returned in the prior
  // `ConverseResponse`. Omit (do not set the field) if there was no prior
  // `ConverseResponse`. If there was a prior `ConverseResponse`, do not omit
  // this field; doing so will end that conversation (and this new request will
  // start a new conversation).
  bytes conversation_state = 1;
}

// The audio containing the assistant's response to the query. Sequential chunks
// of audio data are received in sequential `ConverseResponse` messages.
message AudioOut {
  // *Output-only* The audio data containing the assistant's response to the
  // query. Sequential chunks of audio data are received in sequential
  // `ConverseResponse` messages.
  bytes audio_data = 1;
}

// The semantic result for the user's spoken query.
message ConverseResult {
  // Possible states of the microphone after a `Converse` RPC completes.
  enum MicrophoneMode {
    // No mode specified.
    MICROPHONE_MODE_UNSPECIFIED = 0;

    // The service is not expecting a follow-on question from the user.
    // The microphone should remain off until the user re-activates it.
    CLOSE_MICROPHONE = 1;

    // The service is expecting a follow-on question from the user. The
    // microphone should be re-opened when the `AudioOut` playback completes
    // (by starting a new `Converse` RPC call to send the new audio).
    DIALOG_FOLLOW_ON = 2;
  }

  // *Output-only* The recognized transcript of what the user said.
  string spoken_request_text = 1;

  // *Output-only* The text of the assistant's spoken response. This is only
  // returned for an IFTTT action.
  string spoken_response_text = 2;

  // *Output-only* State information for subsequent `ConverseRequest`. This
  // value should be saved in the client and returned in the
  // `conversation_state` with the next `ConverseRequest`. (The client does not
  // need to interpret or otherwise use this value.) There is no need to save
  // this information across device restarts.
  bytes conversation_state = 3;

  // *Output-only* Specifies the mode of the microphone after this `Converse`
  // RPC is processed.
  MicrophoneMode microphone_mode = 4;

  // *Output-only* Updated volume level. The value will be 0 or omitted
  // (indicating no change) unless a voice command such as "Increase the volume"
  // or "Set volume level 4" was recognized, in which case the value will be
  // between 1 and 100 (corresponding to the new volume level of 1% to 100%).
  // Typically, a client should use this volume level when playing the
  // `audio_out` data, and retain this value as the current volume level and
  // supply it in the `AudioOutConfig` of the next `ConverseRequest`. (Some
  // clients may also implement other ways to allow the current volume level to
  // be changed, for example, by providing a knob that the user can turn.)
  int32 volume_percentage = 5;
}

// The top-level message sent by the client. Clients must send at least two, and
// typically numerous `ConverseRequest` messages. The first message must
// contain a `config` message and must not contain `audio_in` data. All
// subsequent messages must contain `audio_in` data and must not contain a
// `config` message.
message ConverseRequest {
  // Exactly one of these fields must be specified in each `ConverseRequest`.
  oneof converse_request {
    // The `config` message provides information to the recognizer that
    // specifies how to process the request.
    // The first `ConverseRequest` message must contain a `config` message.
    ConverseConfig config = 1;

    // The audio data to be recognized. Sequential chunks of audio data are sent
    // in sequential `ConverseRequest` messages. The first `ConverseRequest`
    // message must not contain `audio_in` data and all subsequent
    // `ConverseRequest` messages must contain `audio_in` data. The audio bytes
    // must be encoded as specified in `AudioInConfig`.
    // Audio must be sent at approximately real-time (16000 samples per second).
    // An error will be returned if audio is sent significantly faster or
    // slower.
    bytes audio_in = 2;
  }
}

// The top-level message received by the client. A series of one or more
// `ConverseResponse` messages are streamed back to the client.
message ConverseResponse {
  // Indicates the type of event.
  enum EventType {
    // No event specified.
    EVENT_TYPE_UNSPECIFIED = 0;

    // This event indicates that the server has detected the end of the user's
    // speech utterance and expects no additional speech. Therefore, the server
    // will not process additional audio (although it may subsequently return
    // additional results). The client should stop sending additional audio
    // data, half-close the gRPC connection, and wait for any additional results
    // until the server closes the gRPC connection.
    END_OF_UTTERANCE = 1;
  }

  // Exactly one of these fields will be populated in each `ConverseResponse`.
  oneof converse_response {
    // *Output-only* If set, returns a [google.rpc.Status][google.rpc.Status]
    // message that specifies the error for the operation. If an error occurs
    // during processing, this message will be set and there will be no further
    // messages sent.
    google.rpc.Status error = 1;

    // *Output-only* Indicates the type of event.
    EventType event_type = 2;

    // *Output-only* The audio containing the assistant's response to the query.
    AudioOut audio_out = 3;

    // *Output-only* The semantic result for the user's spoken query.
    ConverseResult result = 5;
  }
}