MagpieTTS Multilingual 357M
Freehand’s MagpieTTS Multilingual 357M profile supports the v2602 checkpoint through NeMo-Speech.cpp v0.1.0. It generates spoken audio from text with a choice of voices and languages.
Set up speech
Section titled “Set up speech”Follow Local speech with MagpieTTS. NeMo can load Magpie alongside your transcription model. Both share the runtime’s Start and Stop controls.
Follow the NeMo connection guide and enable Text to speech on the connection. Open Text to speech → Settings → Speech, choose that connection and its loaded speech model, then select MagpieTTS Multilingual 357M as the model profile.
Turn on Enable text to speech, choose a voice and language, and Save. Choose Preview to hear your current edits without saving them. Enter text in the composer and choose Speak to use the saved settings. These actions request audio; opening settings and refreshing metadata do not.
Choose a voice and language
Section titled “Choose a voice and language”| Choice | Available values |
|---|---|
| Voices | John, Sofia, Aria, Jason, Leo; each supports all checkpoint languages |
| Default voice | default uses the server’s configured speaker |
| Languages | English, Spanish, German, French, Italian, Vietnamese, Hindi, Mandarin Chinese, Japanese |
| Server default language | Uses server configuration; does not request automatic detection |
Refresh voices reads the loaded model’s metadata. After a successful refresh, Freehand limits language choices to those the server advertises. Japanese and Chinese require NeMo builds with their language frontends enabled.
Available controls
Section titled “Available controls”| Capability | Availability |
|---|---|
| Voice and language | Sent with the request |
| Audio | Complete WAV output |
| Speed adjustment, style, cloning, streamed speech | Unavailable |
| Text normalization | Requires server support and grammar assets; managed setup does not install them |
Your speech voice and language are saved with this model’s options. Generated audio stays in memory until cleared, replaced, recording starts, or Freehand quits. You can explicitly save a WAV file from the player.
Checkpoint version
Section titled “Checkpoint version”Managed setup downloads the versioned Magpie model, NanoCodec decoder, and tokenizer assets qualified with NeMo v0.1.0.
See the v2602 model card, NeMo speech API, and language frontend requirements for the qualified server behavior.