FrontierSeptember 1, 2026via SiliconAngle
Phonely launches Alma, a voice AI model trained on over 10 million phone conversations
Why it matters
A specialized frontier model addresses a real gap in voice AI training (phone conversation authenticity vs. text-trained models). Practitioners building voice agents now have a purpose-built alternative; enthusiasts should track whether voice-native training becomes table stakes for the category.
Key signals
- Phonely launches Alma — voice LLM trained on 10+ million real phone conversations
- Differentiator: trained on phone data, not text (claimed to improve naturalness vs. text-trained competitors)
- Model positioning: built specifically for voice agents
- Training dataset size: 10 million conversations (unusually large and specialized corpus)
The hook
Phonely's Alma: first voice LLM trained on 10M real phone calls, not text data. Changes how voice agents sound.
Phonely Ltd. today announced the launch of Alma, a large language artificial intelligence model built for voice agents, trained on over 10 million real phone conversations. The company said that many AI models driving voice agents have been trained on text. This can make conversations sound slightly…