OpenAI shuts down tts-1, tts-1-hd and gpt-4o-mini-tts on January 6, 2027 (OpenAI deprecations page).

Keep your text-to-speech code running. Change one line.

TTS Swap answers the same /v1/audio/speech request your code sends today and returns audio. No rewrite to the Realtime API.

- client = OpenAI()
+ client = OpenAI(base_url="https://tts.deadlinkpatrol.com/v1", api_key="sk-tts-…")

Your client.audio.speech.create(model="tts-1", voice="alloy", input=…) calls stay as they are.

Your key appears on screen right after checkout. Cancel anytime yourself at the Stripe billing page (enter your checkout email). See the terms.

Hear it before you buy

Real output from this service (model tts-1, voice alloy), unedited WAV files:

English
Spanish
French

1,000,000 characters per month per key.
Same request and error shape as OpenAI, so the official openai Python and Node libraries work.
All 6 classic voice names plus the newer ones are accepted.

The voice will not sound like OpenAI's. Audio comes from the open-source MeloTTS model on Cloudflare Workers AI, which has one voice per language. Every voice name (alloy, nova, onyx…) gets that same voice.
Audio comes back as WAV, even when you ask for mp3 (the default). Browsers, ffmpeg and most players read it fine, but if your code feeds the bytes to an MP3-only decoder, convert first. response_format="pcm" gives raw 24 kHz 16-bit mono like OpenAI. opus, aac and flac get a clear 400 error.
speed is accepted but not applied yet.
English by default. Add extra_body={"language":"es"} for es, fr, zh, ja or ko.
Input up to 4,096 characters per request, like OpenAI.
Limits: 20 requests per minute per key, and a shared daily capacity cap for the whole service. Over a limit you get an OpenAI-style 429 and are not charged characters.

Check your monthly usage: GET /v1/usage with your key. Questions: swiftlabs-ops@agentmail.to

TTS Swap by Swift Labs. Not affiliated with OpenAI. · Terms · Privacy · Manage or cancel subscription