HeyGen Voice: cloning, API access and speech costs

Use the TTS character calculator for APIs that publish character prices. Professional clone slots, subscription charges and minute-based credits need their own budget; the calculator does not convert them.

Use an API key belonging to your HeyGen workspace. Create a clone through POST /v3/models/audio/voices, with mode set to instant or professional. Speech requests use an ACTIVE voice owned by that workspace.

Instant cloning uses one recording and analyzes its first three minutes. Professional cloning takes one to ten recordings of the same speaker totaling at least 20 minutes. Clean speech with little background noise is more useful than padding the recording with silence. Professional training runs asynchronously: wait for the voice to become active before requesting speech.

Use your own voice or obtain the speaker's consent and any required written permission. Do not assume that owning an audio file gives you permission to clone its speaker.

Send text to POST /v3/models/audio/tts for a completed recording, or /v3/models/audio/tts/stream for streaming. Authenticate with X-Api-Key, specify the voice ID and use model heygen-voice-1. Each request accepts 1–5,000 characters; each endpoint permits 30 requests per minute per member.

Completed speech returns a URL for a 44.1 kHz mono PCM16 WAV file and its duration. Streaming sends ordered SSE chunks, each containing a complete WAV segment. Decode the segments as audio rather than joining WAV containers byte for byte. The guide lists 39 base languages, including Chinese; test pronunciation and pacing in the language you need. Instant and Professional voices have different control options, so use the controls documented for your clone mode.

HeyGen’s October 9 official post quotes API speech at $30 per million characters, with a 50% discount to $15 per million characters through October 31, 2026, applied automatically. The post does not specify the cutoff timezone.

Clone creation and training have their own conditions:

Instant clone creation is free during preview, but counts against the workspace's clone allowance. This does not make speech generation free.

Professional cloning requires a purchased slot for each voice. Each slot adds five pooled training runs per month, including the initial run; failed training does not consume one.

Professional speech generation costs 0.6 API credits per generated minute. Ten generated minutes therefore use 6 API credits, before clone-slot charges.

These conditions come from the model guide. A slot that lapses can make the voice unavailable for speech until the allowance is restored.

HeyGen's API is a separate pay-as-you-go purchase from its regular subscriptions, and free API credits ended in February 2026. The public pricing table's Starfish speech price is for another model.

The announcement bills characters without naming a clone mode; the guide bills Professional speech by API credits per minute. Their relationship, the credit-to-dollar conversion, API slot price and character-counting rules remain unspecified in these sources. Use the rates shown in your workspace to budget. The $99 monthly Professional subscription add-on is a separate purchase.

Artificial Analysis places HeyGen Voice in the leading group of its controlled-voice leaderboard. On October 10, HeyGen scored 1201 ±16 and Qwen-Audio-3.1-TTS-Plus scored 1187 ±15; their reported 95% intervals overlap. The evaluation uses eight English voices, four US and four UK. Evaluate Chinese voices separately when that is your workload.

Listen for speaker similarity, names, numbers and pauses in your own scripts. Try short samples before producing a full narration. Check whether your workspace is enabled for Professional cloning before buying a slot.

HeyGen Voice supplies narration. If your project also needs still images, see Seedream 5 Flash; that is a separate visual-generation step with its own charges. Browse the model directory for the rest of the workflow.