evOo-local/packages/adapters/src/kokoro-voice.ts

import type {
  AdapterContext,
  AdapterDescriptor,
  SpeechClip,
  VoiceCapabilities,
  VoiceInfo,
  VoiceProvider,
  VoiceSynthesizeRequest,
  VoiceVerifyResult,
} from "@rakazo/adapter-kit";
import { formatVoiceLabel, GERMAN_VOICE_PRESETS } from "@rakazo/core";
import {
  readVoiceAudio,
  readVoiceJson,
  requireOk,
  voiceDeadline,
  voiceHttpError,
} from "./voice-http.js";

/** Placeholder stored in voice credentials — Kokoro needs no paid API key. */
export const KOKORO_LOCAL_TOKEN = "local-kokoro";

const DEFAULT_BASE_URL = "http://127.0.0.1:8880";
const MODEL = "kokoro";

/** Curated presets when the server is down or returns bare voice ids. */
export const KOKORO_FALLBACK_VOICES: VoiceInfo[] = [
  ...GERMAN_VOICE_PRESETS,
  { id: "af_heart", label: "Heart", description: "warm" },
  { id: "af_bella", label: "Bella", description: "expressive" },
  { id: "af_sky", label: "Sky", description: "bright" },
  { id: "af_nova", label: "Nova", description: "energetic" },
  { id: "am_adam", label: "Adam", description: "clear" },
  { id: "am_michael", label: "Michael", description: "steady" },
  { id: "bf_emma", label: "Emma", description: "British" },
  { id: "bf_isabella", label: "Isabella", description: "British" },
  { id: "bm_george", label: "George", description: "British" },
];

const FALLBACK_BY_ID = new Map(KOKORO_FALLBACK_VOICES.map((voice) => [voice.id, voice]));

export function kokoroBaseUrl(): string {
  const raw = process.env.KOKORO_BASE_URL?.trim();
  return raw && raw.length > 0 ? raw.replace(/\/+$/, "") : DEFAULT_BASE_URL;
}

export function kokoroUnreachable(): string {
  return "Lokaler Sprachdienst nicht erreichbar. Prüfe, ob der Sprachserver auf Port 8880 läuft.";
}

export class KokoroVoiceProvider implements VoiceProvider {
  describe(): AdapterDescriptor<VoiceCapabilities> {
    return {
      id: "kokoro",
      contractVersion: "1",
      adapterVersion: "0.1.0",
      capabilities: { catalog: true, synthesize: true, transcribe: false },
    };
  }

  async verify(_apiKey: string, context: AdapterContext): Promise<VoiceVerifyResult> {
    try {
      const res = await fetch(`${kokoroBaseUrl()}/v1/audio/voices`, {
        signal: voiceDeadline(context.signal, 20_000),
      });
      if (res.ok) return { ok: true };
      return {
        ok: false,
        message: voiceHttpError(
          res.status,
          "Local voice",
          "checking the local server",
          await readVoiceJson(res),
        ),
      };
    } catch {
      return { ok: false, message: kokoroUnreachable() };
    }
  }

  async listVoices(_apiKey: string, context: AdapterContext): Promise<VoiceInfo[]> {
    try {
      const res = await fetch(`${kokoroBaseUrl()}/v1/audio/voices`, {
        signal: voiceDeadline(context.signal, 20_000),
      });
      const body = await readVoiceJson(res, { requireValid: res.ok });
      if (!res.ok) {
        throw new Error(voiceHttpError(res.status, "Local voice", "listing voices", body));
      }
      const voices = voicesFrom(body);
      if (voices.length === 0) return [...KOKORO_FALLBACK_VOICES];
      return voices;
    } catch (error) {
      if (error instanceof Error && /listing voices failed|speaking failed/.test(error.message)) {
        throw error;
      }
      throw new Error(kokoroUnreachable());
    }
  }

  async synthesize(request: VoiceSynthesizeRequest, context: AdapterContext): Promise<SpeechClip> {
    const signal = voiceDeadline(request.signal ?? context.signal, 120_000);
    let res: Response;
    try {
      res = await fetch(`${kokoroBaseUrl()}/v1/audio/speech`, {
        method: "POST",
        headers: { "content-type": "application/json" },
        body: JSON.stringify({
          model: MODEL,
          input: request.text,
          voice: request.voiceId,
          response_format: "mp3",
          stream: false,
        }),
        signal,
      });
    } catch {
      throw new Error(kokoroUnreachable());
    }
    await requireOk(res, "Local voice", "speaking");
    return { bytes: await readVoiceAudio(res, signal), mimeType: "audio/mpeg" };
  }
}

function voicesFrom(body: unknown): VoiceInfo[] {
  const raw = extractVoiceIds(body);
  return raw
    .map((id) => {
      const known = FALLBACK_BY_ID.get(id);
      return known ?? { id, label: formatVoiceLabel(id) };
    })
    .filter((voice) => voice.id);
}

function extractVoiceIds(body: unknown): string[] {
  if (!body || typeof body !== "object") return [];
  const record = body as Record<string, unknown>;
  const voices = record.voices;
  if (!Array.isArray(voices)) return [];
  return voices
    .map((entry) => {
      if (typeof entry === "string") return entry.trim();
      if (entry && typeof entry === "object") {
        const voice = entry as Record<string, unknown>;
        return String(voice.id ?? voice.name ?? "").trim();
      }
      return "";
    })
    .filter(Boolean);
}