GraphAI ~Google STT~

GraphAIでGoogle STTを行うエージェントを作る

import { AgentFunction, AgentFunctionInfo } from "graphai";
import { SpeechClient, protos } from "@google-cloud/speech";
import { GraphAINullableText } from "@graphai/agent_utils";

type STTGoogleInputs = {
  stream: Buffer;
  languageCode?: string;
  sampleRateHertz?: number;
  encoding?: "FLAC" | "ENCODING_UNSPECIFIED" | "LINEAR16" | "MULAW" | "AMR" | "AMR_WB" | "OGG_OPUS" | "SPEEX_WITH_HEADER_BYTE" | "MP3" | "WEBM_OPUS" | null | undefined;
  throwError?: boolean;
};

type STTGoogleConfig = {
  apiKey?: string;
  verbose?: boolean;
};

type STTGoogleParams = STTGoogleInputs & STTGoogleConfig;

type STTGoogleResult = GraphAINullableText

export const sttGoogleAgent: AgentFunction<STTGoogleParams, STTGoogleResult, STTGoogleInputs, STTGoogleConfig> = async ({ params, namedInputs }) => {
  const { stream, languageCode, sampleRateHertz, encoding, throwError } = { ...params, ...namedInputs };
  try {
    const speechClient = new SpeechClient();
    const audioBytes = stream instanceof Buffer ? stream.toString("base64") : stream;

    const audio = {
      content: audioBytes,
    };

    const config = {
      languageCode: languageCode || "ja-JP",
      sampleRateHertz: sampleRateHertz || 48000,
      encoding: encoding || "FLAC",
    };

    const request = {
      audio: audio,
      config: config,
    };

    // Detects speech in the audio file
    const [response] = await speechClient.recognize(request);
    const transcription = response.results
      ? response.results
          .map((result) => 
            result.alternatives && result.alternatives.length > 0
              ? result.alternatives[0].transcript
              : "error"
          )
          .join("\n")
      : "error";

    return {
      text: transcription,
    };
  } catch (e) {
    if (throwError) {
      console.error(e);
      throw new Error("STT Google Error");
    }
    return {
      error: e,
    };
  }
};

const sttGoogleAgentInfo: AgentFunctionInfo = {
  name: "sttGoogleAgent",
  agent: sttGoogleAgent,
  mock: sttGoogleAgent,
  samples: [
    {
      inputs: { stream: Buffer.from("dummy audio data") },
      params: {
        languageCode: "en-US",
        sampleRateHertz: 16000,
        encoding: "LINEAR16",
      },
      result: {
        text: "This is a sample transcription.",
      },
    },
  ],
  description: "Google Speech-to-Text agent",
  category: ["voice"],
  author: "Takuya Matsuda",
  repository: "https://github.com/your-repo/graphai-agents/",
  license: "MIT",
};

export default sttGoogleAgentInfo;