Google has announced Gemini 3.5 Transcribe, an AI model designed to improve voice input by editing out hesitations and corrections for more polished text output. The model is reportedly 70% faster and more accurate than its predecessor, handling 85 languages and up to three speakers in pre-recorded audio.
This advancement in AI-powered speech-to-text technology promises to make voice input more efficient and user-friendly across Google's ecosystem, potentially improving productivity for users who rely on voice for communication and content creation.
Google has introduced Gemini 3.5 Transcribe, a new AI model focused on enhancing speech-to-text capabilities. This model aims to produce more polished transcriptions by automatically editing out verbal fillers such as "ums" and "uhs," as well as self-corrections.
According to Google, Gemini 3.5 Transcribe offers significant performance improvements over its predecessor, Chirp 3. The company states the new model is approximately 70% faster in converting voice to final transcribed text. Additionally, the live-speech error rate has been reduced to 5.5 percent, an improvement from Chirp 3's 7.32 percent.
The AI model is also designed to better understand user intent, capable of removing hesitations and correcting spoken errors in real-time. It can also incorporate custom vocabulary for specialized jargon. The technology supports 85 languages and can process pre-recorded audio featuring up to three speakers.
While the model offers enhanced accuracy and speed, a potential drawback is the AI's alteration of the original spoken words. This may not be suitable for all situations where verbatim transcription is required.