Speech AI and voice synthesis
Voice agents are available today, but the speed of one part tells you little about the speed of the whole conversation. Deepgram separates transcription, model, tool and speech-generation delays from end-to-end response time. Buyers need completed-task accuracy, interruption handling, escalation and the full cost of a call. A sub-100ms component does not prove reliable conversations or that whole service jobs can be replaced.
Research snapshot · updated Nov 12, 2026
State of the art (2026)
Voice agents are on sale, and the speed figures suppliers quote usually cover one part of the system. ElevenLabs says its Flash model takes about 75 milliseconds to generate speech from a short input, before network and application delays are added. A full voice agent also spends time on speech recognition, the language model and audio playback, so the reply a caller hears takes longer. Buyers need the whole response time, the share of tasks completed and the cost per call.
Core thesis
Voice synthesis crossed the uncanny valley years ago; the live question in 2026 is deployment, not naturalness. Real-time speech-to-speech is the default architecture – OpenAI's gpt-realtime reached GA in August 2025, Cartesia's Sonic-3.5 ships sub-100ms synthesis, and ElevenLabs raised a $500M Series D at an $11B valuation in February 2026. Voice agents now handle tier-1 contact-centre work end to end. The same capability cuts both ways: deepfake voice fraud (a CFO-clone authorised a $25M wire transfer on a live call) is driving authentication mandates, and AI music generation faces unresolved IP rulings. The structural growth is real; regulation and fraud liability are the live limits on scale.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
9 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
4 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
93 players · 6 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
6 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
8 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 18
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
8 updates
Thesis changelog
When our view changed, and why
Change our mind
3 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on Speech AI and voice synthesis has changed. All of it is in CanaryIQ.