Pushing the frontiers of audio generation

Our pioneering speech technology applied sciences are serving to individuals around the globe work together with extra pure, conversational and intuitive digital assistants and AI instruments.

Speech is central to human connection. It helps individuals around the globe trade info and concepts, categorical feelings and create mutual understanding. As our expertise constructed for producing pure, dynamic voices continues to enhance, we’re unlocking richer, extra participating digital experiences.

Over the previous few years, we’ve been pushing the frontiers of audio technology, growing fashions that may create prime quality, pure speech from a spread of inputs, like textual content, tempo controls and explicit voices. This expertise powers single-speaker audio in lots of Google merchandise and experiments — together with Gemini Live, Project Astra, Journey Voices and YouTube’s auto dubbing — and helps individuals around the globe work together with extra pure, conversational and intuitive digital assistants and AI instruments.

Working along with companions throughout Google, we just lately helped develop two new options that may generate long-form, multi-speaker dialogue for making advanced content material extra accessible:

• NotebookLM Audio Overviews turns uploaded paperwork into participating and full of life dialogue. With one click on, two AI hosts summarize person materials, make connections between subjects and banter forwards and backwards.
• Illuminate creates formal AI-generated discussions about analysis papers to assist make information extra accessible and digestible.

Right here, we offer an outline of our newest speech technology analysis underpinning all of those merchandise and experimental instruments.

Pioneering strategies for audio technology

For years, we have been investing in audio technology analysis and exploring new methods for producing extra pure dialogue in our merchandise and experimental instruments. In our earlier analysis on SoundStorm, we first demonstrated the power to generate 30-second segments of pure dialogue between a number of audio system.

This prolonged our earlier work, SoundStream and AudioLM, which allowed us to use many text-based language modeling strategies to the issue of audio technology.

SoundStream is a neural audio codec that effectively compresses and decompresses an audio enter, with out compromising its high quality. As a part of the coaching course of, SoundStream learns methods to map audio to a spread of acoustic tokens. These tokens seize the entire info wanted to reconstruct the audio with excessive constancy, together with properties akin to prosody and timbre.

AudioLM treats audio technology as a language modeling activity to supply the acoustic tokens of codecs like SoundStream. In consequence, the AudioLM framework makes no assumptions concerning the sort or make-up of the audio being generated, and might flexibly deal with quite a lot of sounds while not having architectural changes — making it candidate for modeling multi-speaker dialogues.

Constructing upon this analysis, our newest speech technology expertise can produce 2 minutes of dialogue, with improved naturalness, speaker consistency and acoustic high quality, when given a script of dialogue and speaker flip markers. The mannequin additionally performs this activity in underneath 3 seconds on a single Tensor Processing Unit (TPU) v5e chip, in a single inference move. This implies it generates audio over 40-times quicker than actual time.

Scaling our audio technology fashions

Scaling our single-speaker technology fashions to multi-speaker fashions then grew to become a matter of information and mannequin capability. To assist our newest speech technology mannequin produce longer speech segments, we created an much more environment friendly speech codec for compressing audio right into a sequence of tokens, in as little as 600 bits per second, with out compromising the standard of its output.

The tokens produced by our codec have a hierarchical construction and are grouped by time frames. The primary tokens inside a bunch seize phonetic and prosodic info, whereas the final tokens encode superb acoustic particulars.

Even with our new speech codec, producing a 2-minute dialogue requires producing over 5000 tokens. To mannequin these lengthy sequences, we developed a specialised Transformer structure that may effectively deal with hierarchies of knowledge, matching the construction of our acoustic tokens.

With this system, we are able to effectively generate acoustic tokens that correspond to the dialogue, inside a single autoregressive inference move. As soon as generated, these tokens will be decoded again into an audio waveform utilizing our speech codec.

To show our mannequin methods to generate sensible exchanges between a number of audio system, we pretrained it on a whole bunch of hundreds of hours of speech information. Then we finetuned it on a a lot smaller dataset of dialogue with excessive acoustic high quality and exact speaker annotations, consisting of unscripted conversations from various voice actors and sensible disfluencies — the “umm”s and “aah”s of actual dialog. This step taught the mannequin methods to reliably swap between audio system throughout a generated dialogue and to output solely studio high quality audio with sensible pauses, tone and timing.

Consistent with our AI Principles and our dedication to growing and deploying AI applied sciences responsibly, we’re incorporating our SynthID expertise to watermark non-transient AI-generated audio content material from these fashions, to assist safeguard towards the potential misuse of this expertise.

New speech experiences forward

We’re now targeted on bettering our mannequin’s fluency, acoustic high quality and including extra fine-grained controls for options, like prosody, whereas exploring how greatest to mix these advances with different modalities, akin to video.

The potential purposes for superior speech technology are huge, particularly when mixed with our Gemini household of fashions. From enhancing studying experiences to creating content material extra universally accessible, we’re excited to proceed pushing the boundaries of what’s attainable with voice-based applied sciences.