egaki 0.5.0 → 0.7.0

package/dist/cli/speech-catalog.js

@@ -0,0 +1,195 @@
+// Speech model catalog — single source of truth for all TTS models.
+// Mirrors the image/video catalog pattern: every model's provider, pricing,
+// features, and default voices live here. Other modules derive their data
+// from this catalog.
+//
+// Pricing sources:
+//   OpenAI:      https://platform.openai.com/docs/pricing (per 1M characters)
+//   ElevenLabs:  https://elevenlabs.io/pricing (per 1K characters, varies by plan)
+//   Cartesia:    https://cartesia.ai/pricing (plan-based, ~1 credit per character)
+// ─── shared fragments ────────────────────────────────────────────────────────
+const openaiSpeechBase = {
+    provider: 'openai',
+    strategy: 'speech',
+};
+const elevenlabsSpeechBase = {
+    provider: 'elevenlabs',
+    strategy: 'speech',
+};
+const cartesiaSpeechBase = {
+    provider: 'cartesia',
+    strategy: 'speech',
+};
+// ─── catalog ─────────────────────────────────────────────────────────────────
+export const SPEECH_CATALOG = [
+    // ── OpenAI ─────────────────────────────────────────────────────────────
+    {
+        id: 'tts-1',
+        name: 'TTS-1 (Standard)',
+        description: 'OpenAI standard text-to-speech. Low latency, good for real-time use.',
+        ...openaiSpeechBase,
+        released: '2023-11',
+        cost: { type: 'per-character', perMillionChars: 15 },
+        features: {
+            outputFormats: ['mp3', 'opus', 'aac', 'flac', 'wav', 'pcm'],
+            instructions: false,
+            speed: true,
+            language: false,
+            maxChars: 4096,
+            defaultVoices: ['alloy', 'ash', 'coral', 'echo', 'fable', 'onyx', 'nova', 'sage', 'shimmer'],
+        },
+    },
+    {
+        id: 'tts-1-hd',
+        name: 'TTS-1 HD',
+        description: 'OpenAI high-definition text-to-speech. Higher fidelity audio.',
+        ...openaiSpeechBase,
+        released: '2023-11',
+        cost: { type: 'per-character', perMillionChars: 30 },
+        features: {
+            outputFormats: ['mp3', 'opus', 'aac', 'flac', 'wav', 'pcm'],
+            instructions: false,
+            speed: true,
+            language: false,
+            maxChars: 4096,
+            defaultVoices: ['alloy', 'ash', 'coral', 'echo', 'fable', 'onyx', 'nova', 'sage', 'shimmer'],
+        },
+    },
+    {
+        id: 'gpt-4o-mini-tts',
+        name: 'GPT-4o Mini TTS',
+        description: 'OpenAI GPT-4o Mini with text-to-speech. Supports style instructions ' +
+            'for expressive speech control.',
+        ...openaiSpeechBase,
+        released: '2025-03',
+        cost: { type: 'per-character', perMillionChars: 12 },
+        features: {
+            outputFormats: ['mp3', 'opus', 'aac', 'flac', 'wav', 'pcm'],
+            instructions: true,
+            speed: true,
+            language: false,
+            maxChars: 2000,
+            defaultVoices: [
+                'alloy', 'ash', 'ballad', 'cedar', 'coral', 'echo',
+                'fable', 'marin', 'nova', 'onyx', 'sage', 'shimmer', 'verse',
+            ],
+        },
+    },
+    // ── ElevenLabs ─────────────────────────────────────────────────────────
+    {
+        id: 'eleven_v3',
+        name: 'ElevenLabs v3',
+        description: 'Latest ElevenLabs model with best quality and expressiveness.',
+        ...elevenlabsSpeechBase,
+        released: '2025-06',
+        cost: { type: 'per-character', perMillionChars: 100 },
+        features: {
+            outputFormats: ['mp3_44100_128', 'mp3_22050_32', 'pcm_16000', 'pcm_22050', 'pcm_24000', 'pcm_44100', 'ulaw_8000'],
+            instructions: false,
+            speed: false,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [],
+        },
+    },
+    {
+        id: 'eleven_multilingual_v2',
+        name: 'ElevenLabs Multilingual v2',
+        description: 'ElevenLabs multilingual model. Supports 29 languages.',
+        ...elevenlabsSpeechBase,
+        released: '2024-01',
+        cost: { type: 'per-character', perMillionChars: 100 },
+        features: {
+            outputFormats: ['mp3_44100_128', 'mp3_22050_32', 'pcm_16000', 'pcm_22050', 'pcm_24000', 'pcm_44100', 'ulaw_8000'],
+            instructions: false,
+            speed: false,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [],
+        },
+    },
+    {
+        id: 'eleven_flash_v2_5',
+        name: 'ElevenLabs Flash v2.5',
+        description: 'ElevenLabs low-latency model optimized for speed.',
+        ...elevenlabsSpeechBase,
+        released: '2024-08',
+        cost: { type: 'per-character', perMillionChars: 50 },
+        features: {
+            outputFormats: ['mp3_44100_128', 'mp3_22050_32', 'pcm_16000', 'pcm_22050', 'pcm_24000', 'pcm_44100', 'ulaw_8000'],
+            instructions: false,
+            speed: false,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [],
+        },
+    },
+    {
+        id: 'eleven_turbo_v2_5',
+        name: 'ElevenLabs Turbo v2.5',
+        description: 'ElevenLabs turbo model. Balanced quality and speed.',
+        ...elevenlabsSpeechBase,
+        released: '2024-06',
+        cost: { type: 'per-character', perMillionChars: 50 },
+        features: {
+            outputFormats: ['mp3_44100_128', 'mp3_22050_32', 'pcm_16000', 'pcm_22050', 'pcm_24000', 'pcm_44100', 'ulaw_8000'],
+            instructions: false,
+            speed: false,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [],
+        },
+    },
+    // ── Cartesia ───────────────────────────────────────────────────────────
+    // Pricing is plan-based (~1 credit per character). The per-million rate
+    // below is estimated from the Startup plan ($49/1.25M credits ≈ $39/M).
+    // See https://cartesia.ai/pricing for current plans.
+    {
+        id: 'sonic-3.5',
+        name: 'Cartesia Sonic 3.5',
+        description: 'Fastest, most natural Cartesia TTS. #1 for naturalness, sub-90ms latency, ' +
+            '42 languages. Recommended for production.',
+        ...cartesiaSpeechBase,
+        released: '2026-05',
+        cost: { type: 'per-character', perMillionChars: 39 },
+        features: {
+            outputFormats: ['mp3', 'wav', 'raw'],
+            instructions: false,
+            speed: true,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [
+                'f786b574-daa5-4673-aa0c-cbe3e8534c02', // Katie (en-US Female)
+                'db6b0ed5-d5d3-463d-ae85-518a07d3c2b4', // Skylar (en-US Female)
+                'a5136bf9-224c-4d76-b823-52bd5efcffcc', // Jameson (en-US Male)
+                '62ae83ad-4f6a-430b-af41-a9bede9286ca', // Gemma (en-GB Female)
+                'ef191366-f52f-447a-a398-ed8c0f2943a1', // Archie (en-GB Male)
+            ],
+        },
+    },
+    {
+        id: 'sonic-3',
+        name: 'Cartesia Sonic 3',
+        description: 'Previous generation Cartesia TTS. Still high quality with pronunciation ' +
+            'dictionaries support.',
+        ...cartesiaSpeechBase,
+        released: '2025-01',
+        cost: { type: 'per-character', perMillionChars: 39 },
+        features: {
+            outputFormats: ['mp3', 'wav', 'raw'],
+            instructions: false,
+            speed: true,
+            language: true,
+            maxChars: 0,
+            defaultVoices: [
+                'f786b574-daa5-4673-aa0c-cbe3e8534c02', // Katie (en-US Female)
+                'db6b0ed5-d5d3-463d-ae85-518a07d3c2b4', // Skylar (en-US Female)
+                'a5136bf9-224c-4d76-b823-52bd5efcffcc', // Jameson (en-US Male)
+            ],
+        },
+    },
+];
+// ─── lookup helpers ──────────────────────────────────────────────────────────
+export function findSpeechModel(modelId) {
+    return SPEECH_CATALOG.find((m) => m.id === modelId);
+}

package/dist/cli/speech-generate.d.ts

@@ -0,0 +1,86 @@
+import type { GeneratedFile } from './generate.js';
+import type { WordTimestamp } from './transcription-generate.js';
+export type { WordTimestamp };
+export interface SpeechProviderResult {
+    audio: Uint8Array;
+    mediaType: string;
+    /** Word-level timestamps, if the provider supports them. */
+    timestamps?: WordTimestamp[];
+}
+export interface SpeechProviderOptions {
+    text: string;
+    modelId: string;
+    voice?: string;
+    outputFormat?: string;
+    instructions?: string;
+    speed?: number;
+    language?: string;
+    abortSignal?: AbortSignal;
+}
+export interface SpeechProvider {
+    generate(options: SpeechProviderOptions): Promise<SpeechProviderResult>;
+}
+/** Speech model IDs from the catalog. Accepts arbitrary strings too. */
+export type SpeechModelId = 'tts-1' | 'tts-1-hd' | 'gpt-4o-mini-tts' | 'eleven_v3' | 'eleven_multilingual_v2' | 'eleven_flash_v2_5' | 'eleven_turbo_v2_5' | 'sonic-3.5' | 'sonic-3' | (string & {});
+/** Built-in voice presets across providers. Accepts arbitrary strings too. */
+export type SpeechVoice = 'alloy' | 'ash' | 'coral' | 'echo' | 'fable' | 'onyx' | 'nova' | 'sage' | 'shimmer' | 'ballad' | 'cedar' | 'marin' | 'verse' | (string & {});
+/** Audio output formats. */
+export type AudioOutputFormat = 'mp3' | 'opus' | 'aac' | 'flac' | 'wav' | 'pcm' | 'raw' | (string & {});
+export interface GenerateSpeechOptions {
+    /** The text to convert to speech. */
+    text: string;
+    /** Speech model ID. Defaults to DEFAULT_SPEECH_MODEL ('tts-1') if omitted. */
+    model?: SpeechModelId;
+    /** Voice ID or name (provider-specific). */
+    voice?: SpeechVoice;
+    /** Output audio format: mp3, wav, pcm, opus, aac, flac, etc. */
+    outputFormat?: AudioOutputFormat;
+    /** Style instructions for models that support it (e.g. gpt-4o-mini-tts). */
+    instructions?: string;
+    /**
+     * Playback speed multiplier. `1.0` is normal speed; values below slow
+     * down, values above speed up.
+     *
+     * Accepted ranges vary by provider:
+     *
+     * | Provider    | Range       | Notes                                                        |
+     * |-------------|-------------|--------------------------------------------------------------|
+     * | OpenAI      | 0.25 – 4.0  | Works with `tts-1`, `tts-1-hd`, `gpt-4o-mini-tts`.          |
+     * | ElevenLabs  | 0.25 – 4.0  | Passed as `voice_settings.speed`. Extreme values degrade     |
+     * |             |             | quality. Flash/turbo models may ignore it.                   |
+     * | Cartesia    | 0.6 – 1.5   | Passed as `generation_config.speed`. Sonic models treat it   |
+     * |             |             | as guidance, not an exact multiplier.                        |
+     *
+     * @default 1.0
+     */
+    speed?: number;
+    /** ISO 639-1 language code (e.g. "en", "es", "fr"). */
+    language?: string;
+    /** Abort signal for cancelling the request. */
+    abortSignal?: AbortSignal;
+}
+export interface GenerateSpeechResult {
+    audio: GeneratedFile;
+    model: string;
+    cost: number | null;
+    /** Word-level timestamps, if the provider supports them (ElevenLabs, Cartesia). */
+    timestamps?: WordTimestamp[];
+}
+export declare function calculateSpeechCost(cost: {
+    type: 'per-character';
+    perMillionChars: number;
+}, textLength: number): number;
+/**
+ * Generate speech audio from text. Auto-detects which provider to use
+ * based on model ID. Shares credentials and subscription with the CLI.
+ * Returns word-level timestamps when the provider supports them.
+ */
+export declare function generateSpeechUncached(opts: GenerateSpeechOptions): Promise<Error | GenerateSpeechResult>;
+export interface CachedSpeechResult {
+    src: string;
+    timestamps?: WordTimestamp[];
+}
+export declare const generateSpeech: import("./cached-generate.js").CachedGenerateFn<GenerateSpeechOptions & {
+    seed?: number;
+}, CachedSpeechResult>;
+//# sourceMappingURL=speech-generate.d.ts.map

package/dist/cli/speech-generate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"speech-generate.d.ts","sourceRoot":"","sources":["../../src/cli/speech-generate.ts"],"names":[],"mappings":"AAwBA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAA;AAClD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAA;AAEhE,YAAY,EAAE,aAAa,EAAE,CAAA;AAK7B,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,UAAU,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,4DAA4D;IAC5D,UAAU,CAAC,EAAE,aAAa,EAAE,CAAA;CAC7B;AAED,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,WAAW,CAAC,EAAE,WAAW,CAAA;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,oBAAoB,CAAC,CAAA;CACxE;AAID,wEAAwE;AACxE,MAAM,MAAM,aAAa,GACrB,OAAO,GAAG,UAAU,GAAG,iBAAiB,GACxC,WAAW,GAAG,wBAAwB,GAAG,mBAAmB,GAAG,mBAAmB,GAClF,WAAW,GAAG,SAAS,GACvB,CAAC,MAAM,GAAG,EAAE,CAAC,CAAA;AAEjB,8EAA8E;AAC9E,MAAM,MAAM,WAAW,GACnB,OAAO,GAAG,KAAK,GAAG,OAAO,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,QAAQ,GAAG,OAAO,GAAG,OAAO,GAAG,OAAO,GAC5H,CAAC,MAAM,GAAG,EAAE,CAAC,CAAA;AAEjB,4BAA4B;AAC5B,MAAM,MAAM,iBAAiB,GAAG,KAAK,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC,CAAA;AAIvG,MAAM,WAAW,qBAAqB;IACpC,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAA;IACZ,8EAA8E;IAC9E,KAAK,CAAC,EAAE,aAAa,CAAA;IACrB,4CAA4C;IAC5C,KAAK,CAAC,EAAE,WAAW,CAAA;IACnB,gEAAgE;IAChE,YAAY,CAAC,EAAE,iBAAiB,CAAA;IAChC,4EAA4E;IAC5E,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB;;;;;;;;;;;;;;;OAeG;IACH,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,+CAA+C;IAC/C,WAAW,CAAC,EAAE,WAAW,CAAA;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,aAAa,CAAA;IACpB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,GAAG,IAAI,CAAA;IACnB,mFAAmF;IACnF,UAAU,CAAC,EAAE,aAAa,EAAE,CAAA;CAC7B;AAID,wBAAgB,mBAAmB,CACjC,IAAI,EAAE;IAAE,IAAI,EAAE,eAAe,CAAC;IAAC,eAAe,EAAE,MAAM,CAAA;CAAE,EACxD,UAAU,EAAE,MAAM,GACjB,MAAM,CAER;AA+CD;;;;GAIG;AACH,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,qBAAqB,GAAG,OAAO,CAAC,KAAK,GAAG,oBAAoB,CAAC,CAmC/G;AAwBD,MAAM,WAAW,kBAAkB;IACjC,GAAG,EAAE,MAAM,CAAA;IACX,UAAU,CAAC,EAAE,aAAa,EAAE,CAAA;CAC7B;AAQD,eAAO,MAAM,cAAc;WAAmD,MAAM;sBA8BlF,CAAA"}

package/dist/cli/speech-generate.js

@@ -0,0 +1,156 @@
+// Programmatic API for egaki speech generation.
+// Returns Error | Result (errore style) instead of process.exit/console.error.
+//
+// Two variants:
+//   - `generateSpeech(opts)` — cached, writes to public/generated/audio/, returns { src }
+//   - `generateSpeechUncached(opts)` — raw, returns bytes in GenerateSpeechResult
+//
+// Uses our own SpeechProvider interface instead of the Vercel AI SDK's
+// experimental_generateSpeech. This lets providers return word-level timestamps
+// alongside audio (ElevenLabs /with-timestamps, Cartesia /tts/sse).
+//
+// Usage:
+//   import { generateSpeech, generateSpeechUncached } from 'egaki/generate'
+//   const cached = await generateSpeech({ text: 'Hello world' })
+//   if (!(cached instanceof Error)) cached.src // '/generated/audio/hello-world-a1b2c3d4.mp3'
+//   const raw = await generateSpeechUncached({ text: 'Hello world' })
+//   if (!(raw instanceof Error)) raw.audio.uint8Array // raw bytes
+//   if (raw.timestamps) raw.timestamps[0] // { word: 'Hello', startSecond: 0, endSecond: 0.4 }
+import { injectCredentialsToEnv } from './credentials.js';
+import { getSpeechModelConfig, getSpeechProvider, DEFAULT_SPEECH_MODEL, } from './speech-models.js';
+// ─── cost calculation ────────────────────────────────────────────────────────
+export function calculateSpeechCost(cost, textLength) {
+    return (textLength / 1_000_000) * cost.perMillionChars;
+}
+// ─── media type inference ────────────────────────────────────────────────────
+function inferAudioMediaType(format) {
+    switch (format) {
+        case 'mp3':
+            return 'audio/mpeg';
+        case 'wav':
+            return 'audio/wav';
+        case 'pcm':
+        case 'raw':
+            return 'audio/pcm';
+        case 'opus':
+            return 'audio/opus';
+        case 'aac':
+            return 'audio/aac';
+        case 'flac':
+            return 'audio/flac';
+        default:
+            return 'audio/mpeg';
+    }
+}
+function inferMediaTypeFromBytes(audio, requestedFormat) {
+    // Check magic bytes first
+    if (audio.length >= 3 && audio[0] === 0x49 && audio[1] === 0x44 && audio[2] === 0x33) {
+        return 'audio/mpeg'; // ID3 header (MP3)
+    }
+    if (audio.length >= 2 && audio[0] === 0xff && (audio[1] & 0xe0) === 0xe0) {
+        return 'audio/mpeg'; // MP3 sync word
+    }
+    if (audio.length >= 4 && audio[0] === 0x52 && audio[1] === 0x49 && audio[2] === 0x46 && audio[3] === 0x46) {
+        return 'audio/wav'; // RIFF header
+    }
+    if (audio.length >= 4 && audio[0] === 0x4f && audio[1] === 0x67 && audio[2] === 0x67 && audio[3] === 0x53) {
+        return 'audio/ogg'; // OggS header (could be opus)
+    }
+    if (audio.length >= 4 && audio[0] === 0x66 && audio[1] === 0x4c && audio[2] === 0x61 && audio[3] === 0x43) {
+        return 'audio/flac'; // fLaC header
+    }
+    // Fall back to requested format
+    return inferAudioMediaType(requestedFormat);
+}
+// ─── generateSpeech ──────────────────────────────────────────────────────────
+/**
+ * Generate speech audio from text. Auto-detects which provider to use
+ * based on model ID. Shares credentials and subscription with the CLI.
+ * Returns word-level timestamps when the provider supports them.
+ */
+export async function generateSpeechUncached(opts) {
+    injectCredentialsToEnv();
+    const model = opts.model ?? DEFAULT_SPEECH_MODEL;
+    const config = getSpeechModelConfig(model);
+    if (config instanceof Error)
+        return config;
+    const provider = getSpeechProvider(config.provider);
+    if (provider instanceof Error)
+        return provider;
+    let result;
+    try {
+        result = await provider.generate({
+            text: opts.text,
+            modelId: model,
+            voice: opts.voice,
+            outputFormat: opts.outputFormat,
+            instructions: opts.instructions,
+            speed: opts.speed,
+            language: opts.language,
+            abortSignal: opts.abortSignal,
+        });
+    }
+    catch (err) {
+        return err instanceof Error ? err : new Error(String(err));
+    }
+    const mediaType = result.mediaType || inferMediaTypeFromBytes(result.audio, opts.outputFormat);
+    const cost = calculateSpeechCost(config.cost, opts.text.length);
+    return {
+        audio: { uint8Array: result.audio, mediaType },
+        model,
+        cost,
+        timestamps: result.timestamps,
+    };
+}
+// ─── cached generateSpeech ──────────────────────────────────────────────────
+// Writes audio to public/generated/audio/ and a sidecar .timestamps.json
+// next to it when the provider returns word timestamps. On cache hit,
+// the sidecar is read back to populate the result.
+//
+// Uses cachedGenerate with postWrite to write the timestamps sidecar after
+// the audio file, and deserialize to read it back on cache hit.
+import fs from 'node:fs';
+import path from 'node:path';
+import { cachedGenerate } from './cached-generate.js';
+import { extensionFromMediaType } from './cache-utils.js';
+/** Compute the sidecar timestamps path for an audio file.
+ *  Stored in a `timestamps/` subdirectory to avoid cache key collision:
+ *  findCachedFile() matches any filename containing the hash, so a sidecar
+ *  like `foo-hash.mp3.timestamps.json` in the same directory would be
+ *  mistakenly returned as the "cached audio" file. */
+function timestampsPathFor(filePath) {
+    return path.join(path.dirname(filePath), 'timestamps', path.basename(filePath) + '.json');
+}
+export const generateSpeech = cachedGenerate({
+    namespace: 'audio',
+    prefixFrom: (p) => p.text,
+    modelFrom: (p) => p.model,
+    generate: async (params) => {
+        const result = await generateSpeechUncached(params);
+        if (result instanceof Error)
+            throw result;
+        return { audio: result.audio, timestamps: result.timestamps };
+    },
+    serialize: (output) => ({
+        bytes: output.audio.uint8Array,
+        extension: extensionFromMediaType(output.audio.mediaType),
+    }),
+    postWrite: (filePath, output) => {
+        if (output.timestamps?.length) {
+            const tsPath = timestampsPathFor(filePath);
+            fs.mkdirSync(path.dirname(tsPath), { recursive: true });
+            fs.writeFileSync(tsPath, JSON.stringify(output.timestamps, null, 2));
+        }
+    },
+    deserialize: ({ urlPath, filePath }) => {
+        const tsPath = timestampsPathFor(filePath);
+        let timestamps;
+        try {
+            if (fs.existsSync(tsPath)) {
+                timestamps = JSON.parse(fs.readFileSync(tsPath, 'utf-8'));
+            }
+        }
+        catch { /* non-fatal: timestamps are optional */ }
+        return { src: urlPath, timestamps };
+    },
+});

package/dist/cli/speech-models.d.ts

@@ -0,0 +1,8 @@
+import type { SpeechProvider } from './speech-generate.js';
+import { type SpeechModelEntry } from './speech-catalog.js';
+export { type SpeechModelEntry };
+export declare const DEFAULT_SPEECH_MODEL = "tts-1";
+export declare function getSpeechModelConfig(modelId: string): Error | SpeechModelEntry;
+export declare function ensureSpeechProviderKey(providerName: string): Error | undefined;
+export declare function getSpeechProvider(providerName: string): Error | SpeechProvider;
+//# sourceMappingURL=speech-models.d.ts.map

package/dist/cli/speech-models.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"speech-models.d.ts","sourceRoot":"","sources":["../../src/cli/speech-models.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAA;AAM1D,OAAO,EAGL,KAAK,gBAAgB,EACtB,MAAM,qBAAqB,CAAA;AAE5B,OAAO,EAAE,KAAK,gBAAgB,EAAE,CAAA;AAEhC,eAAO,MAAM,oBAAoB,UAAU,CAAA;AAqC3C,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,MAAM,GAAG,KAAK,GAAG,gBAAgB,CAM9E;AAgCD,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,GAAG,SAAS,CAY/E;AAUD,wBAAgB,iBAAiB,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,GAAG,cAAc,CA0B9E"}

package/dist/cli/speech-models.js

@@ -0,0 +1,109 @@
+// Speech model registry for egaki.
+// Maps speech model IDs to their provider and SpeechProvider factory.
+//
+// Provider resolution priority (same as image/video):
+//   1. Direct provider key → direct provider implementation
+//   2. Egaki API key → route through egaki gateway (future)
+//   3. No key → error with subscription recommendation
+import pc from 'picocolors';
+import { PROVIDERS, injectCredentialsToEnv, } from './credentials.js';
+import { ValidationError } from './models.js';
+import { SPEECH_CATALOG, findSpeechModel, } from './speech-catalog.js';
+export {};
+export const DEFAULT_SPEECH_MODEL = 'tts-1';
+const SPEECH_PROVIDER_FACTORIES = {
+    openai: async () => {
+        const { createOpenAISpeechProvider } = await import('./openai-speech-provider.js');
+        return createOpenAISpeechProvider();
+    },
+    elevenlabs: async () => {
+        const { createElevenLabsSpeechProvider } = await import('./elevenlabs-speech-provider.js');
+        return createElevenLabsSpeechProvider();
+    },
+    cartesia: async () => {
+        const { createCartesiaSpeechProvider } = await import('./cartesia-provider.js');
+        return createCartesiaSpeechProvider();
+    },
+};
+// ─── key checking ────────────────────────────────────────────────────────────
+function hasDirectProviderKey(providerName) {
+    const info = PROVIDERS[providerName];
+    if (!info)
+        return false;
+    return Boolean(process.env[info.envVar]);
+}
+function hasEgakiKey() {
+    const info = PROVIDERS['egaki'];
+    if (!info)
+        return false;
+    return Boolean(process.env[info.envVar]);
+}
+// ─── config lookup ───────────────────────────────────────────────────────────
+export function getSpeechModelConfig(modelId) {
+    const entry = findSpeechModel(modelId);
+    if (!entry) {
+        return new ValidationError(`Unknown speech model: ${modelId}`);
+    }
+    return entry;
+}
+function resolveAuthSource(providerName) {
+    const info = PROVIDERS[providerName];
+    if (info && process.env[info.envVar]) {
+        return { type: 'api-key', label: info.envVar };
+    }
+    if (hasEgakiKey()) {
+        return { type: 'egaki-gateway' };
+    }
+    return { type: 'api-key', label: 'unknown' };
+}
+function logAuthSource(source) {
+    switch (source.type) {
+        case 'api-key':
+            console.error(pc.dim(`Auth: ${source.label}`));
+            break;
+        case 'egaki-gateway':
+            console.error(pc.dim('Auth: Egaki subscription (gateway)'));
+            break;
+    }
+}
+// ─── ensure key ──────────────────────────────────────────────────────────────
+export function ensureSpeechProviderKey(providerName) {
+    if (hasDirectProviderKey(providerName))
+        return undefined;
+    // Speech models don't go through the gateway yet (AI Gateway doesn't
+    // support speech routing). Direct provider key is required.
+    const info = PROVIDERS[providerName];
+    const label = info?.label || providerName;
+    return new ValidationError(`Missing API key for ${label}. ` +
+        `Run: egaki login --provider ${providerName} --key <key>. ` +
+        (info ? info.hint : ''));
+}
+// ─── provider cache ──────────────────────────────────────────────────────────
+/** Global cache of provider instances, keyed by provider name. */
+const providerCache = globalThis.__egakiSpeechProviderCache ??= new Map();
+// ─── public factory ──────────────────────────────────────────────────────────
+export function getSpeechProvider(providerName) {
+    injectCredentialsToEnv();
+    const keyError = ensureSpeechProviderKey(providerName);
+    if (keyError)
+        return keyError;
+    const authSource = resolveAuthSource(providerName);
+    logAuthSource(authSource);
+    const factory = SPEECH_PROVIDER_FACTORIES[providerName];
+    if (!factory) {
+        return new Error(`Speech generation is not supported for provider: ${providerName}`);
+    }
+    // Return a lazy provider: the factory is async but we want getSpeechProvider
+    // to be sync for ergonomic use in generateSpeechUncached. The provider promise
+    // is cached globally per provider name so the factory runs once across all calls.
+    return {
+        async generate(options) {
+            const cached = providerCache.get(providerName);
+            if (cached)
+                return (await cached).generate(options);
+            const promise = factory();
+            providerCache.set(providerName, promise);
+            return (await promise).generate(options);
+        },
+    };
+}

package/dist/cli/speech-provider.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=speech-provider.test.d.ts.map

package/dist/cli/speech-provider.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"speech-provider.test.d.ts","sourceRoot":"","sources":["../../src/cli/speech-provider.test.ts"],"names":[],"mappings":""}