Drama 3 upgraded weights and is the #1 open-weight model on Artificial Analysis Speech Arenas, ranking #1 Provider Voice with +38 Elo. Controlled Voice ranking is #1 open-weight with +22 Elo; it also tops the Pronunciation Robustness benchmark.
Key points
Provider Voice: 1 open-weight, #6 overall, +38 Elo over Preview
Controlled Voice: 1 open-weight, +22 Elo over Preview
Sources
Bluejay engineering moves evals to Fargate, cuts cost 78%
Bluejay moved its evaluation workloads from Lambda to Fargate, achieving a 78% cost reduction and 25% latency improvement. The company also shipped end-to-end email simulations, Digital Twins, and Ask Bluejay in Slack, and launched the MISTT speech-to-text benchmark.
Key points
Cost reduction: 78% cost reduction moving evals from Lambda to Fargate
Latency reduction: 25% latency reduction moving evals from Lambda to Fargate
Sources
Eleven SFX v3 launch – richer sound from text prompt
Eleven SFX v3 lets you create richer, more precise sound effects from a single text prompt. It adds improved music and vocals and is now available inside ElevenCreative and ElevenAPI.
Key points
SFX v3 creates richer, more precise sound effects from a single text prompt.
Available inside ElevenCreative and ElevenAPI.
Sources
Audio-to-MIDI adds lyric transcription feature
Audio-to-MIDI now transcribes lyrics along with the notes. The tool can export vocal sheets with lyrics as PDF or MusicXML.
Key points
Audio-to-MIDI transcribes lyrics along with the notes.
Export options include vocal sheets with lyrics in PDF or MusicXML.
Sources
Soniox STT and TTS support 60+ languages, mix mid-sentence
Soniox STT and TTS support 60+ languages and can handle language switches mid-sentence, with TTS mixing languages in responses. Developers get consistent multilingual coverage for both speech recognition and synthesis from a single provider.
Key points
Languages: 60+ languages supported by both STT and TTS
Switches: STT recognizes language switches mid-sentence; TTS naturally mixes languages in its responses
Sources
seedance 2.5 and kling 4 added to h3 music video generator
seedance 2.5, seedance 2 mini/pro/fast and kling 4 are now live alongside h3 in the music video generator. Hybrid presets claim to give the best bang for your buck and scenes can be retaken with an upper model in an h3 project.
Key points
seedance 2.5, seedance 2 mini/pro/fast and kling 4 now live alongside h3 in the music video generator.
hybrid presets help you get the best bang for your buck.
you can retake any scene with an upper model in an h3 project.
Sources
GradiumAI Chrome extension released: read and translate
GradiumAI Chrome extension released: read and translate any content. It also can transcribe content and was built for fun.
Key points
Chrome extension released last week
Can read, transcribe and translate any content
Sources
PostHog analytics use and SoundBoost customer story
PostHog is used for analytics and decision making since day one, and SoundBoost was featured in a PostHog customer story.
Key points
Using PostHog for analytics and decision making since day one
SoundBoost featured in a PostHog customer story
Sources
- Original post
- posthog.com - Linked in the post