The long tail of languages
Search 646 languages by name and pick the exact one, down to regional forms such as Tunisian or Hijazi Arabic.
Musa Sona 600 reaches the languages most voice tools leave out: Urdu, Swahili, Indonesian, Tamil, Vietnamese and hundreds more, including regional dialects. It is our largest engine and our fastest on a GPU.
“ایک چھوٹے سے خوبصورت گاؤں میں بٹو نام کا ایک لکڑھارا رہتا تھا۔ وہ روزانہ جنگل جاتا، لکڑیاں کاٹتا اور انہیں بازار میں فروخت کر کے اپنا گزر بسر کرتا۔”
“السلام علیکم دوستو۔ آج ہم سیکھیں گے کہ گھر بیٹھے اپنا نیا کام کیسے شروع کریں۔”
Peace be upon you, friends. Today we’ll learn how to start your new work from home.
“Hello friends. Today we will learn how to start new work from home.”
“Bonjour à tous. Aujourd'hui, nous allons apprendre à commencer un nouveau travail depuis la maison.”
Hello everyone. Today we’ll learn how to start new work from home.
“வணக்கம் நண்பர்களே. இன்று வீட்டிலிருந்தே புதிய வேலையை எப்படி தொடங்குவது என்று கற்றுக்கொள்வோம்.”
Hello friends. Today we’ll learn how to start new work from home.
“Xin chào các bạn. Hôm nay chúng ta sẽ học cách bắt đầu một công việc mới ngay tại nhà.”
Hello everyone. Today we’ll learn how to start a new job right at home.
“Halo teman-teman. Hari ini kita akan belajar cara memulai pekerjaan baru dari rumah.”
Hello friends. Today we’ll learn how to start new work from home.
“Habari zenu marafiki. Leo tutajifunza jinsi ya kuanza kazi mpya ukiwa nyumbani.”
Hello friends. Today we’ll learn how to start new work from home.
Musa Sona 600 on a GPU, cloned from the Urdu sample on the left. The input voices are AI-generated. Only clone voices you own or have permission to use.
Search 646 languages by name and pick the exact one, down to regional forms such as Tunisian or Hijazi Arabic.
On a mid-range NVIDIA card, Sona made 60 seconds of speech in 9 seconds.
Choose between faster and best quality, and decide how closely the result follows the reference voice.
[laughter], [sigh] and four more bring natural reactions to a script.
Choose a tag from the Sound menu, or type it where it should happen.
[laughter][sigh][surprise-ah][surprise-oh][question-ah][dissatisfaction-hnn] A few of the 646. Search the full list inside the app.
Sona needs a strong GPU and a transcript of your sample. Results vary between languages and voices, and a voice with one accent can struggle in a distant language. Test yours before relying on it.
On Windows, an NVIDIA GPU with 8 GB of memory and the free GPU pack. On a Mac, Apple Silicon with 16 GB of memory. It is not offered on CPU-only PCs, and the Models page tells you if your computer qualifies before you download anything.
Sona learns the voice from a recording and its exact words. Saved voices already have a transcript in My Voices. For a new clip you can type the words, or let the app transcribe it with your own Groq key.
Only if you choose Groq, an optional online service used with your own key. Generation itself is local. Type the transcript yourself and nothing leaves your computer.
No. We word-checked French, Hindi, Urdu, Spanish, German, Swahili, Japanese and English, with good results. Across 646 languages quality varies, so test your language with your voice first.
Install the free version and test it with your own script. One license covers every engine.