Ce que Deepgram a publié en 2026

Voice Agents Meet the Phone Line: What It Really Takes to Ship in Production

Deepgram and Twilio announced a joint technical webinar on August 19, 2026, focused on deploying voice agents on phone lines. The session highlights challenges like latency, codecs, voicemail, and com…

Deepgram CLI 0.3.0 adds Flux TTS support, fixes Flux STT streaming, and enforces exit codes

Deepgram released CLI 0.3.0 with full Flux TTS support via `dg speak`, fixes for Flux STT streaming in `dg listen`, and a breaking change enforcing exit codes (0=success, 1=error, 2=user interrupt). T…

Pricing change detected for Deepgram

Pricing updated for Deepgram: - Pay-As-You-Go: promotion changed — Free $200 Credit then pay-as-you-go - Developers & Startups: $4000/yr → Custom pricing - Developers & Startups: allowance changed fro…

Voice AI Infrastructure Leader Deepgram to Establish APAC Headquarters in Singapore Following Strategic Investment from EDBI

Deepgram will open its Asia-Pacific headquarters in Singapore, appointing Sriram Ved as VP and GM for the region. The move follows a 96% YoY increase in APAC API requests and a $130M Series C extensio…

Deepgram Flux TTS for Developers: What Ships at General Availability

Deepgram’s Flux TTS API reached general availability on a new /v2/speak endpoint, replacing the older /v1/speak. The update introduces turn-based streaming for real-time voice agents, batch processing…

Deepgram Nova-3 adds Afrikaans and Georgian with improved models for 7 languages

Deepgram expanded Nova-3 with two new languages (Afrikaans and Georgian) and improved monolingual models for seven existing languages (Russian, Hungarian, Macedonian, Slovak, Slovenian, and Urdu). The…

Why TTS Latency Grows With Response Length and the Architecture That Stops It

The article explains how TTS latency scales with response length due to autoregressive generation dependencies, contrasting it with parallel/non-autoregressive models. It highlights that streaming run…

Deepgram adds Numerals support for Bulgarian, Cantonese, Malay, and Korean

Deepgram expanded its Numerals feature to support four new languages—Bulgarian, Chinese (Cantonese, Traditional), Malay, and Korean—using monolingual models. The feature converts spoken numbers into d…

Pricing updated for Deepgram: - Pay-As-You-Go: promotion changed — $200 free credit - New tier: Developers & Startups ($4000/yr) - New tier: Growing Applications ($4000/yr) - Removed: Custom tier

Introducing Flux STT: Conversational Speech Recognition to Solve the Biggest Problem in Voice Agents-Interruptions

Deepgram introduced Flux STT, the first speech recognition model designed specifically for conversational voice AI, not just transcription. Unlike traditional ASR systems, Flux STT integrates turn det…

Introducing Flux TTS: Conversation-Native Text to Speech for Real Time Voice Agents

Deepgram introduced Flux TTS, the first conversation-native text-to-speech model designed for real-time voice agents, addressing failures of traditional TTS in live interactions. The model maintains c…

Deepgram Flux TTS adds precise interruption handling for voice agents

Deepgram’s Flux TTS now supports precise interruption handling via a new `Interrupt` message that stops audio playback locally and sends a `playback_offset` to the server. The server responds with `Sp…