AI
ElevenLabs ships v4 speech models in 90 languages
ElevenLabs released two new speech models on Monday: v4, and v4 Turbo, which has lower latency (less waiting before sound comes out) and is meant for voice agents. Turbo can start generating audio as the language model begins its reply.
Languages go from 70 to 90, with better Japanese, Brazilian Portuguese, Mandarin and Cantonese. Voice cloning needs about 10 seconds of audio. The inline expression tags introduced in v3 can now be stacked, so you can direct a performance in sequence. No pricing was given.
TechCrunch also reports ElevenLabs' run rate (yearly revenue pace) rose from about $330M to over $600M, and that it is reportedly in talks at a $22B valuation. Reported, not confirmed.
On your desk: The Stacy Show's puppet voices are the obvious test, especially the stackable expression tags for comic timing. Travis's coaching clips could reach other languages; have a native speaker check the results.
Read the original at TechCrunch: ElevenLabs v4 speech model →