Yesterday, Google in its open-source blog announced the Android open source speech recognition transcription tool - Live Transcribe speech engine (Live Transcribe Speech Engine), which aims to voice or text conversations in real-time transcription is also able to assist the hearing impaired.
Live Transcribe is Google in February this year launched an Android application that Google's speech recognition by the most advanced Cloud Speech API provided. However, depending on the cloud introduces some complexity, changing network connections, data costs and delays, etc., bring some robustness tests. Therefore, Google put it out of open source, developers hope to further develop and build on the existing basis.
Cloud Speech API does not yet support unlimited audio stream, the current team has taken some steps to address this problem, such as turning off before reaching out and restart the streaming request, which will effectively reduce the amount of text session loss.
Unlimited streaming audio brings a great challenge. In many countries, the network data is very expensive and poor Internet where bandwidth may be limited. Live Transcribe Speech Engine team audio codec large number of experiments, and eventually without affecting the accuracy of the data used in an amount reduced by 10-fold.
In addition, because it is to provide real-time voice transcription, the transcribed text will be entered as the voice of the constantly changing, reducing the delay is naturally very necessary. The engine can greatly reduce the delay rate, thanks to its custom Opus encoder.
In addition, it is worth mentioning that, Live Transcribe supports more than 70 languages, and can be based on automatic speech recognition languages, including Chinese.
