Hello everyone,
I’m a software engineer currently building an AI-powered application and would appreciate some technical guidance from the Google AI team or the community.
Our goal is not voice cloning, dubbing, translation, or text-to-speech.
Instead, we are trying to build a workflow that preserves the original speaker while improving only the pronunciation.
The intended pipeline is:
- A user uploads a prerecorded video.
- Our application extracts the original audio from the video.
- If necessary, we also provide a transcript of the spoken content to help the AI correctly interpret words that may be difficult to recognize because of a strong foreign accent.
- The AI processes the speech while preserving:
- the speaker’s original voice identity,
- emotions,
- speaking style,
- pacing,
- pauses,
- timing,
- and overall delivery.
- The only desired change is improved pronunciation and reduced foreign accent.
- The processed audio is merged back into the original video and returned to the user.
The person should still sound exactly like themselves — same personality, same timing, same emotions — only with more natural pronunciation.
Over the past few weeks, I’ve contacted several speech AI companies (Krisp, Sanas, Deepgram, Speechmatics, Rime, and others), but I haven’t received a clear technical answer regarding whether this exact workflow is currently supported.
My questions are:
- Can this workflow be implemented using Gemini or other Google AI speech technologies?
- Would this require combining multiple models, or is there already a speech-to-speech approach that supports it?
- Which Google APIs or models would you recommend for preserving voice identity while improving pronunciation?
- If this is not currently possible, is it something Google AI is actively researching or planning to support in the future?
Any technical guidance or recommendations would be greatly appreciated.
Thank you very much!