Speech
Turn text into speech in the playground with preset voices, voices you describe, or your own cloned voice, and download the clips.
The Speech playground (/app/playground?mode=speech) turns text into speech. You pick a voice, type the text and press Speak; each clip is billed to your balance like an API call. To do the same from code, see Text to speech and the Audio endpoints.
On a phone, the model is in the dropdown at the top and the voice settings open with Options.
Pick a model#
Choose in Model. Under it you see the price per minute of audio and what the model is good at:
qwen3-tts: describe a voice or clone one, style instructions, 10 languages.zonos2: close voice clones and emotions, 40+ languages.kokoro-82m: 54 preset voices in 8 languages, the lowest price, $0.008 per minute; no cloning.drama-3: acting directions in [brackets], cloning, 83 languages, $0.09 per minute; MP3 at normal speed only.
The first two cost $0.036 per minute, billed by the second. See Models for current prices.
The models run on GPUs that sleep when nobody uses them. The badge under the model shows Warm (answers right away), Waking up (usually ready within the range shown) or Asleep (your first request waits while it starts; after that it is fast). Only a request wakes it, not opening the page. If a request takes longer than a few seconds, a note says the model may be waking up: the first one after a quiet spell can take up to a minute or two. See Hosted models.
Choose a voice#
Voice has three tabs.
Preset#
A list of the model's built-in voices, each with a short description and its native language. qwen3-tts has nine (for example serena, ryan, ono_anna); zonos2 has american_female, american_male and british_female. Every voice speaks every language its model supports, best in its native one.
Describe#
On qwen3-tts only. Describe the voice you want: age, gender, timbre, pace, mood, accent. The chips Radio host, Bright young woman, Old pirate and Bedtime story fill in examples. A designed voice comes out a little different each time; to keep one you like, save the clip as a voice.
My voices#
Your saved voices. Each row has a play button for the sample, its name and transcript; click the row to use it. The cross, then Delete?, removes it. A saved voice works with both models.
To add one, press + Add a voice:
- Upload an audio file, or press Record to use your microphone. Recording stops when you press Stop or after 20 seconds. Use 3 to 15 seconds of one person speaking clearly, without music.
- Give it a Name.
- Optionally type what the recording says. With it,
qwen3-ttsclones more closely. - Tick the consent box: you confirm the voice is yours or you have the speaker's permission to clone it, and that you are responsible for how you use it. Don't clone voices to deceive or impersonate people.
- Press Save voice.
Saved voices are free and private to your account. Limits are in Clone a voice.
Style and emotion#
- Style (
qwen3-tts, with a preset or saved voice): how to say it, such as pace, mood or a whisper. Type your own or pick a chip: Warm and friendly, Excited, a little fast, Slow and calm, Whispering, Sad, close to tears, Like a news anchor. Click a chip again to clear it. With Describe, the description takes this role instead. - Emotion (
zonos2): Neutral, Happy, Sad, Angry or Surprised. The voice stays the same; only the delivery changes.
Language, format and speed#
- Language: Auto detects it from the text, or pick one of the model's languages.
- Format: MP3, WAV, OPUS, FLAC or AAC. OPUS downloads as an
.oggfile. - Speed: 0.5× to 2× with the slider. It changes the pace, not the pitch. You pay for the seconds you get back.
Speak#
Type up to 4,096 characters in the text box; the counter shows how many you have used. The English example, Chinese example and Japanese example chips fill in sample text. Press Speak or Ctrl+Enter (Cmd+Enter on a Mac). Next to the button you see roughly how long the audio will be and what it will cost.
Your clips#
Each clip appears above the earlier ones with the voice, the model, its length and its cost, and a player to listen to it.
- Download saves the file in the format you chose.
- Save as a voice asks for a name and saves the clip to My voices, then selects it. Use it to keep a voice you designed. A clip of up to 15 seconds is saved with its text as the transcript.
Clips stay in the list until you close or reload the page; download the ones you want to keep. Every request is also listed in your logs.
Do it from code#
View code shows the cURL request for your current settings. The fields are described in Text to speech and the Audio endpoints reference.