Anyone who has yelled “representative!” at a phone bot knows the problem. Voice agents either sound robotic, or they sound great but can’t be controlled. Rime says it has a fix.
On Wednesday, the voice AI company launched Prism, which it describes as the first composable speech-to-speech model. Rime says Prism responds to callers in about 300 milliseconds, makes tool calls in parallel with speech and exposes a text layer that businesses can run guardrails over before the model speaks.
The trade-off it’s trying to fix
In a blog post, Rime’s Lily Clifford lays out the dilemma. Traditional “cascade” voice systems convert speech to text, reason over the text and convert back. They’re controllable, but they throw away the hesitation, frustration or laughter in a caller’s voice. Fused speech-to-speech models from big labs keep that signal, but you can’t inspect what they’re about to say.
Prism is trained end to end to listen, write what it’s going to say and keep hearing the user while it speaks. The text acts as a checkpoint where rules can read every word, plus support for pronouncing names and account terms correctly.
The benchmark claims
Rime says it ran Prism on the harness Artificial Analysis uses for its speech-to-speech leaderboard, without a reasoning step. On τ-Voice, which simulates customer-service calls in airline, retail and telecom, Prism scored 56.6, which Rime says beats realtime models from OpenAI and Google that don’t reason first. In the telecom domain, it scored 82.5 versus 45.6 for GPT-Live-1, with 0.57 seconds to first audio.
Rime also says Prism is the first speech-to-speech model anyone can fine-tune, and it’s building reinforcement learning and supervised fine-tuning APIs with beta customers. In one test, a Prism trained on 10 hours of ordering-agent calls from Revmo confirmed with callers before placing 94% of orders, compared with 57% for the OpenAI model Revmo uses today.
These are company-reported results, so buyers will want to test them on their own calls.
Why it matters
Voice is where AI customer service gets real, and enterprises want both natural conversation and compliance controls. Models that let teams inspect and fine-tune what an agent says could speed up adoption in regulated contact centers.
For contact center platforms, CX consultancies and VoIP providers, it means buyers are actively re-evaluating their voice stack. That’s a good moment to reach CX and operations leaders with clear, proof-backed messaging about AI voice agents.
Sources: Rime. Written by the xRapid Media LLP team.
Selling to the IT market? xRapid Media LLP is a High Tech B2B Marketing company. We help technology companies reach decision-makers with verified B2B data, account-based campaigns, email and LinkedIn outreach and appointment setting. Call +91 9767196309 or book a free B2B marketing consultation.