Phonon by GradiumAI is a local TTS model with int8 weights of 44MB. It runs 26× real-time on an iPhone in Safari and is under 40MB when Brotli compressed. The model is open source under Apache-2.0 and timestamps every word.
Key points
int8 weights size 44MB
26× real-time performance on iPhone Safari
compressed size under 40MB Brotli
open source Apache-2.0 license, every word timestamped
Sources
RiversidedotFM records first podcast episode, 11 clips ready
RiversidedotFM was used to record the first episode of a new podcast. The editor reported a plan ready in 5 minutes, and Clipper pulled 11 clips ready to be cut a minute later.
Key points
Editor messaged that the plan was ready 5 minutes after finishing.
Clipper pulled 11 clips ready to be cut one minute later.
Sources
Pika H3 Flex faster video model with audio issues
Pika H3 Flex is reported as about 1.8× faster than Fal H3 Max for 768p DiT denoising and nearly 2× faster for longer video generations. The post also notes that its audio lacks fidelity and sync, making the overall video feel off.
Key points
1.8× faster DiT denoising than Fal H3 Max at 768p.
Nearly 2× speed for longer video generations.
Audio lacks fidelity and sync, affecting video quality.
Sources
Hausa as spoken in Kano and Katsina dataset with 23 speakers
Hausa as it is spoken in Kano and Katsina dataset includes 23 speakers from Nigeria. Unscripted answers were recorded on their own phones and each clip has a human-validated transcript.
Key points
23 speakers in Nigeria, 16 from Kano and Katsina.
Every clip carries a human-validated transcript.
Sources
ElevenLabs skills 12 tools reach 110k downloads
ElevenLabs skills package now has 110k downloads and includes 12 skills for coding agents. It adds Text to Speech, Speech to Text, ElevenAgents, music, dubbing and more. Install with npx skills add elevenlabs/skills.
Key points
Downloads: 110k downloads
Skill set: 12 skills include Text to Speech, Speech to Text, ElevenAgents, music, dubbing
Sources
Voice AI in automotive market map lists 43 companies across 5 segments
Voice AI in automotive market map lists 43 companies across 5 segments. The post groups them into Dealership platform, Dealership service, Repair and parts, EV charging, and Dealership sales. Bluejay is working with some of the listed companies.
Key points
Companies: 43 companies listed.
Segments: 5 segments defined.
Sources
- Original post
- diago.ai - Linked in the post
Vande Mataram – AI Remix creates new song in 90 seconds
Vande Mataram, written in 1875, was rebuilt in about 90 seconds using Soundverse v9 to produce a new track titled "Radiant Motherland". An AI Remix then kept the melody and generated cinematic orchestral anthem and Indian classical fusion versions.
Key points
Rebuilt in about 90 seconds.
New song titled "Radiant Motherland" with its own cover art.
Sources
Soundverse adds ChatGPT integration via custom MCP server
Soundverse now works inside ChatGPT by adding it as a custom MCP server and connecting your account. Ask for a song in plain English and receive the full track in the chat, saved to your Soundverse Library.
Key points
Server URL and client ID (jt7cp9hv4m9idkuyv2gt0) are required to add the connector.
A ChatGPT plan that allows custom MCP servers is needed.
Sources
- Original post
- Soundverse — AI Music, Video & Voice Generator - Linked page
