Skip to main content
Profy Expert provides two voice capabilities: Text-to-Speech (TTS) that converts text into natural-sounding human voice, and Automatic Speech Recognition (ASR) that transcribes audio files into text.

​
Two Core Capabilities

Text-to-Speech (TTS)

Convert text content into natural, fluent speech audio. Supports multiple voice styles, ideal for audio content creation.

Speech Recognition (ASR)

Transcribe speech from audio files into text. Supports long audio, ideal for meeting notes and content organization.

​
Text-to-Speech (TTS)

​
How to Use

Tell the Expert which text to voice. The audio arrives as an inline file card in the conversation, ready to play or download:
Pick a speech model first: TTS reads the speech model from the model selector. With none selected it fails outright (“No speech model selected”) — it does not fall back to your chat model. Batch generation uses tts_batch, submitting several segments at once with a concurrency cap of 20 — the right shape for voicing a long script section by section.

​
Voice Selection

You can specify your preferred voice type in your request. Different models offer different voice options, typically including:
  • Male / Female
  • Young / Mature
  • Gentle / Powerful
  • News broadcast / Casual conversation
Not sure which voice to pick? Have Expert read a sample sentence in different voices, then choose your favorite before generating the full content.

​
Use Cases

Convert written manuscripts into audio content — quickly produce sample clips for audiobooks or audio courses.
Generate narration voiceovers for video content, pairing with the video generation capability.
Convert documents and web content into speech for visually impaired users or for consuming information on the go.
Quickly convert written content into audio format for publishing on podcast or audio platforms.

​
Speech Recognition (ASR)

​
How to Use

1

Upload an Audio File

Drag and drop the audio file you want to transcribe (e.g., MP3, WAV, M4A) into the chat, or click the attachment button to upload.
2

Request Transcription

Tell Expert what you need, such as “transcribe this recording into text” or “extract the key points from this meeting recording.”
3

Get the Results

Expert will return the transcribed text content. You can then ask Expert to format the text, extract summaries, or translate it into other languages.

​
Use Cases

Upload meeting recordings and have Expert transcribe and organize structured meeting minutes, including topics, decisions, and action items.
Convert interview recordings into text transcripts. Expert can also help distinguish between different speakers.
After transcribing recordings, have Expert extract key points, generate mind maps, or organize study notes.
After transcription, you can directly have Expert translate the content — enabling a “listen in one language, read in another” workflow.
After transcription is complete, you can directly ask Expert to further process the text — summarize key points, extract critical information, generate action item lists, and more — without switching tools.

​
Usage Tips

  • Audio quality directly affects recognition accuracy. Use clear recordings with minimal background noise whenever possible
  • Very noisy environment recordings may result in transcription gaps or errors
  • Technical jargon or uncommon proper nouns may require manual correction after transcription

​
Billing

Whether TTS settles per character or per 10,000 characters depends on the pricing unit configured for that model in the admin panel — the database config is the single source of truth, and the unit reported by the runtime is only a fallback hint. Switching speech models can therefore switch the billing basis for identical text.Counting follows the MiniMax convention: one Chinese character counts as 2, one Latin letter as 1. A 500-character Chinese script is about 1,000 billable characters, so don’t estimate from visible character count.
Voice input in the chat box is free and does not follow the ASR billing above — POST /api/audio/transcribe has no billing hook. What costs credits is the turn that follows, once the transcribed text enters the conversation as tokens.The ASR billing above applies when an Expert actively calls speech recognition to process an audio file as part of a task (for example, “turn this meeting recording into minutes”). These are two different entry points; do not conflate them.
Per-model rates are operational data (read the model selector); the algorithms are fixed and documented in Billing Formulas.

​
Failures and fixes

Supported audio formats are wav, mp3, m4a and webm. Uploaded audio is still bound by per-file limits (audio falls under “other”, 512 MB) — see Limits.