File transcription and speech synthesis

Open Audio Playground. Available models follow your current Playground plan and organization; locked models cannot be selected. These requests consume Playground subscription credit, separately from API wallet calls.

Transcribe a file

Choose the file transcription task and a supported model. Upload a WAV, MP3, FLAC, OGG, M4A or AAC file up to 18 MiB. The browser checks its contents rather than trusting the filename. An optional instruction can specify language or output style. Confirm the usage notice and start.

The existing gemini-3.5-transcribe model receives input_audio through /v1/chat/completions; this does not introduce an /audio/transcriptions API. A valid empty result means no speech was detected, not zero usage. Copy or download the transcript before leaving.

Generate speech

Choose speech synthesis, an available TTS model, one of its supported voices, and WAV or PCM. Input is limited to 5,000 characters. WAV can be played in the browser; raw PCM is downloadable 24 kHz, 16-bit mono and needs a compatible player. Playback depends on the browser and output device. Audio is not archived by this page.

The synthesis request uses /v1/audio/speech and a short-lived Key restricted to one speech model. Chat, embedding and Realtime models are rejected on this surface. It does not send the Chat-only max_tokens parameter. The runtime's verified audio output cap and existing request/cumulative budgets remain in force.

Cancellation, privacy and usage

Confirming sends the selected file or text to the model provider. Stop cancels the browser request; already produced usage may still settle. There are no automatic paid retries. The result identifies the actual request organization and provides its exact short-lived Key receipt. Delayed or missing evidence remains pending/unknown, never a confirmed zero.

Results and original media remain in page memory until you leave or replace them. Download results to retain them. The page does not cloud-save audio, transcripts or credentials. Standalone API calls use your API wallet and a long-lived API Key; see Speech API.