cross pond high tech

Researchers trained a neural network to map audio “voiceprints” from one language to another. The results aren’t perfect, but you can sort of hear how Google’s translator was able to retain the voice and tone of the original speaker. It can do this because it converts audio input directly to audio output without any intermediary steps. In contrast, traditional translational systems convert audio into text, translate the text, and then resynthesize the audio, losing the characteristics of the original voice along the way. The new system, dubbed the Translatotron, has three components, all of which look at the speaker’s audio spectrogram—a visual snapshot of the frequencies used when the sound is playing, often called a voiceprint. The first component uses a neural network trained to map the audio spectrogram in the input language to the audio spectrogram in the output language. The second converts the spectrogram into an audio wave that can be played. The third component can then layer the original speaker’s vocal characteristics back into the final audio output. Not only does this approach produce more nuanced translations by retaining important nonverbal cues, but in theory it should also minimize translation error, because it reduces the task to fewer steps. Translatotron is currently a proof of concept. During testing, the researchers trialed the system only with Spanish-to-English translation, which already took a lot of carefully curated training data. But audio outputs like the clip above demonstrate the potential for a commercial system later down the line. You can listen to more of them here.

AI can generate storyboard animations from scripts, spot potholes and cracks in roads, and teach four-legged robots to recover when they fall. But what about adapting one person’s singing style to that of another? Yep — it’s got that down pat, too. In a paper published on the preprint server Arxiv.org (“Unsupervised Singing Voice Conversion“), scientists at Facebook AI Research and Tel Aviv University describe a system that directly converts audio of one singer to the voice of another. All the more impressive, it’s unsupervised, meaning it’s able to perform the conversion from unclassified, unannotated data it hasn’t previously encountered.

The team claims that their model was able to learn to convert between singers from just 5-30 minutes of their singing voices, thanks in part to an innovative training scheme and data augmentation technique.

“[Our approach] could lead, for example, to the ability to free oneself from some of the limitations of one’s own voice,” the paper’s authors wrote. “The proposed network is not conditioned on the text or on the notes [and doesn’t] require parallel training data between the various singers, nor [does it] employ a transcript of the audio to either text … or to musical notes … While existing pitch correction methods … correct local pitch shifts, our work offers flexibility along the other voice