Meta Muse Voice Transcribe is launched for real-time audio transcription. This model developed by Meta Superintelligence Labs supports more than 70 languages and works in 5 Indian languages including Hindi, Tamil, Telugu, Kannada and Malayalam.
Muse Voice Transcribe can convert speech to text while the conversation is taking place instead of waiting for the recording to complete. The model can recognize the voices of different speakers as well as determine the start and end times of their speaking.
According to Meta, Muse Voice Transcribe can differentiate between more than 20 human voices in the same recording. It is capable of processing audio longer than an hour, reducing the need to break long conversations into shorter clips.
The model processes audio in small chunks of approximately 80 milliseconds. Then determine whether enough information has been obtained to transcribe a word accurately. Simple words are processed quickly, while for ambiguous words the model waits for additional audio.
Meta said, Reinforcement Learning has been used in this process. Through this, the aim has been to determine the appropriate time of transcription and reduce errors during speech.
Muse Voice Transcribe – has been trained on over 70 languages. The company tested the model in 25 languages before its initial release. Apart from Hindi, these include Tamil, Telugu, Kannada and Malayalam.
The model can also handle language switching during a conversation. If the language changes even in the middle of the same sentence, the user does not have to manually change the language settings.
Apart from this, the system supports Language, Keyword and Context Biasing. Through this, it helps in identifying the right word according to the context of the conversation.
Meta Muse Voice Transcribe is made available through the Meta Model API. Developers can use it to add features like speech transcription to their applications. It is being used for Meta AI and Muse Code for Mac for dictation.
The API is priced at $3 (approximately ₹300) for 1,000 audio minutes. Thus, the cost of one hour of audio is approximately $0.18 (approximately ₹17).
