@omg-dev/media 0.4.25 → 0.4.26

package/dist/index.mjs

@@ -87,5 +87,53 @@ async function generateVideo(opts) {
 		job
 	};
 }
+/**
+* Text → speech. Submits, polls to completion, and returns a durable CDN URL
+* to the generated mp3. TTS runs in seconds, but in a server route still
+* prefer submit + getJob (or an `on()` handler) — see the media docs.
+*
+*   const { url } = await generateSpeech({ text: "Welcome back!" })
+*/
+async function generateSpeech(opts) {
+	const { text, voiceId, model, input, ...rest } = opts;
+	const job = await generate({
+		model: model ?? "elevenlabs/eleven_turbo_v2_5",
+		input: {
+			text,
+			...voiceId ? { voice_id: voiceId } : {},
+			...input
+		},
+		...rest
+	});
+	const url = job.results?.[0]?.url;
+	if (!url) throw new Error(`media generateSpeech: job ${job.jobId} produced no result`);
+	return {
+		url,
+		job
+	};
+}
+/**
+* Speech → text. Submits, polls to completion, and returns the transcript.
+*
+*   const { text } = await transcribe({ audioUrl, durationSeconds: 42 })
+*/
+async function transcribe(opts) {
+	const { audioUrl, durationSeconds, model, input, ...rest } = opts;
+	const job = await generate({
+		model: model ?? "elevenlabs/scribe_v1",
+		input: {
+			audio_url: audioUrl,
+			duration_seconds: durationSeconds,
+			...input
+		},
+		...rest
+	});
+	const text = job.results?.[0]?.text;
+	if (text == null) throw new Error(`media transcribe: job ${job.jobId} produced no transcript`);
+	return {
+		text,
+		job
+	};
+}
 //#endregion
-export { generate, generateImage, generateVideo, getEndUser, getJob, runWithEndUser, submit };
+export { generate, generateImage, generateSpeech, generateVideo, getEndUser, getJob, runWithEndUser, submit, transcribe };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@omg-dev/media",
-  "version": "0.4.25",
+  "version": "0.4.26",
   "type": "module",
   "exports": {
     ".": {

package/src/index.ts

@@ -52,6 +52,8 @@ export type JobStatus = "queued" | "running" | "succeeded" | "failed"
 export interface ResultRef {
   url: string
   contentType?: string
+  /** Inline textual output (speech-to-text transcripts). No url when set. */
+  text?: string
 }
 
 export interface Job {
@@ -165,3 +167,71 @@ export async function generateVideo(
   if (!url) throw new Error(`media generateVideo: job ${job.jobId} produced no result`)
   return { url, job }
 }
+
+// ── Speech (ElevenLabs) ──────────────────────────────────────────────────────
+
+export interface SpeechOptions extends Omit<GenerateOptions, "model" | "input"> {
+  /** The text to speak. Billed per character (pass-through). */
+  text: string
+  /** ElevenLabs voice id; the platform default voice when omitted. */
+  voiceId?: string
+  /** Curated model id (default "elevenlabs/eleven_turbo_v2_5"). */
+  model?: string
+  /** Extra model params merged into the input (voice_settings, output_format…). */
+  input?: Record<string, unknown>
+}
+
+/**
+ * Text → speech. Submits, polls to completion, and returns a durable CDN URL
+ * to the generated mp3. TTS runs in seconds, but in a server route still
+ * prefer submit + getJob (or an `on()` handler) — see the media docs.
+ *
+ *   const { url } = await generateSpeech({ text: "Welcome back!" })
+ */
+export async function generateSpeech(
+  opts: SpeechOptions,
+): Promise<{ url: string; job: Job }> {
+  const { text, voiceId, model, input, ...rest } = opts
+  const job = await generate({
+    model: model ?? "elevenlabs/eleven_turbo_v2_5",
+    input: { text, ...(voiceId ? { voice_id: voiceId } : {}), ...input },
+    ...rest,
+  })
+  const url = job.results?.[0]?.url
+  if (!url) throw new Error(`media generateSpeech: job ${job.jobId} produced no result`)
+  return { url, job }
+}
+
+export interface TranscribeOptions extends Omit<GenerateOptions, "model" | "input"> {
+  /** Public URL of the audio to transcribe (e.g. an uploaded file's URL). */
+  audioUrl: string
+  /**
+   * Length of the audio in seconds — REQUIRED, it is what you're billed on
+   * ($0.22/audio-hour pass-through, 60s minimum). Understating it undercharges;
+   * read it off the file/recorder before submitting.
+   */
+  durationSeconds: number
+  /** Curated model id (default "elevenlabs/scribe_v1"). */
+  model?: string
+  /** Extra model params merged into the input (language_code…). */
+  input?: Record<string, unknown>
+}
+
+/**
+ * Speech → text. Submits, polls to completion, and returns the transcript.
+ *
+ *   const { text } = await transcribe({ audioUrl, durationSeconds: 42 })
+ */
+export async function transcribe(
+  opts: TranscribeOptions,
+): Promise<{ text: string; job: Job }> {
+  const { audioUrl, durationSeconds, model, input, ...rest } = opts
+  const job = await generate({
+    model: model ?? "elevenlabs/scribe_v1",
+    input: { audio_url: audioUrl, duration_seconds: durationSeconds, ...input },
+    ...rest,
+  })
+  const text = job.results?.[0]?.text
+  if (text == null) throw new Error(`media transcribe: job ${job.jobId} produced no transcript`)
+  return { text, job }
+}