Voice

ElevenLabs released Overdub 2.0 in May 2026, a major update to its voice-cloning and speech-editing tool that lets podcasters, video producers, and audiobook narrators fix spoken mistakes by typing the correction — the AI generates the replacement audio in the creator's own voice and stitches it seamlessly into the original recording. The update ships with an API that enables automated content pipelines: broadcast media workflows can now route flagged segments through Overdub automatically rather than scheduling re-recording sessions with on-air talent. The voice quality at launch is described by early users as indistinguishable from source material in A/B listening tests, which removes the previous bottleneck where obvious audio quality discontinuities required producers to re-record larger surrounding sections rather than isolated words. Overdub 2.0 is available on ElevenLabs' Creator plan at $12 per month and Pro at $24 per month, with a free tier for limited edits. For content creators and audio producers who spend significant time in post-production cleanup, Overdub 2.0 converts a multi-hour fix-and-re-record workflow into a text-editing task measured in minutes.

OpenAI released three new realtime voice models through its API in May 2026: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. The launch moves voice AI from conversation-style chatbots into a new category of voice agents capable of reasoning mid-conversation, calling tools without interrupting the speaker, and recovering from ambiguous or incomplete input without dropping context. GPT-Realtime-Translate is purpose-built for multilingual scenarios — customer support, healthcare intake, educational tutoring — where a live human speaker and an AI need to interact across a language barrier without adding a separate translation step. GPT-Realtime-Whisper focuses on transcription accuracy in noisy environments and medical-grade clarity, directly competing with Deepgram and AssemblyAI in the transcription API market. For developers building voice-first applications in 2026, the trio replaces a multi-vendor stack that previously required separate models for speech recognition, translation, and conversational response — simplifying architecture while improving latency for real-world voice agent use cases.

OpenAI is building toward a voice-first physical hardware device expected to ship in 2027, with a dedicated audio language model as the prerequisite step planned for release earlier in 2026, according to reporting from The Information confirmed by organizational restructuring inside OpenAI. The company has reorganized teams specifically to develop audio-based AI products — a signal that voice interfaces are no longer a secondary feature within ChatGPT but a standalone product category OpenAI is investing in with dedicated engineering and model resources. Former Apple design lead Jony Ive has publicly argued that voice-controlled AI products are likely to be less screen-dependent and more ambient than current smartphone interfaces, and OpenAI's audio hardware roadmap aligns with the broader industry shift toward AI companions and assistants that operate through conversation rather than text. For voice tool developers and text-to-speech vendors building on top of current OpenAI APIs, the in-house audio model and hardware roadmap signals that OpenAI will compete directly in the voice generation and voice interface market it currently enables for third parties.

Alongside its transcription model, Microsoft released a proprietary voice generation engine as part of its MAI model family — marking the first time the company has built an in-house text-to-speech system capable of competing with ElevenLabs, OpenAI's TTS models, and Google's voice synthesis products. The model is designed for the same enterprise workflows where Microsoft's ecosystem already dominates: meeting summaries read aloud, accessibility tools, automated customer communications, and AI-powered assistants embedded in Teams and Outlook. CEO Satya Nadella confirmed that the MAI voice model will eventually power Cortana-adjacent experiences across Microsoft's consumer and enterprise products, though specific integration timelines were not disclosed at launch. The practical implication for enterprise buyers is cost consolidation: companies that currently pay for separate ElevenLabs or third-party voice synthesis licenses for their Microsoft-integrated workflows may be able to drop those contracts once MAI voice is fully embedded in Azure AI services. Microsoft emphasized that the voice model was built with "platform of platforms" philosophy — it will sit alongside Anthropic's Claude and OpenAI's models in Microsoft's Foundry API, giving developers the option to route voice tasks to the Microsoft model while routing reasoning tasks to GPT or Claude. For developers building voice-first applications on Azure, having a competitive in-house model at platform pricing rather than a third-party API rate is a meaningful cost reduction.

Apple is shipping Gemini-powered Siri as part of iOS 26.4, expected to roll out to users the week of April 26. This is the first consumer-facing version of Siri that runs on a frontier AI model — in this case, Google's Gemini — instead of Apple's own voice assistant backend. The impact on everyday use is significant. Gemini brings two capabilities that Siri has never had: deep context handling across a conversation, and tool use. Context handling means Siri can now remember what you said three messages ago and respond coherently. If you ask "What's the weather this weekend?" and then follow up with "What should I pack?", it understands the connection without you repeating yourself. Tool use means Siri can take actions in apps: setting calendar events, drafting messages, searching emails, and pulling up specific documents — all from natural language requests in the same conversation. Apple's own AI features (Writing Tools, Image Playground, and on-device summaries) remain unchanged. Gemini is specifically powering the conversational assistant layer — the part you talk to. Privacy protections: Apple confirmed that Siri queries routed to Gemini follow Google's enterprise privacy commitments, with no training on user conversations by default. This is a notable moment: Apple built the world's most popular AI assistant, watched it fall behind, and chose to partner rather than compete on the model layer. For the hundreds of millions of iPhone users who gave up on Siri years ago, iOS 26.4 is worth trying again.

Google released Flash TTS on April 20—a new text-to-speech model that lets users control speaking style, pace, pitch, and emotional emphasis via natural language instead of preset voice selections. Traditional TTS systems force you to choose from 20–50 pre-built voices; Flash TTS lets you say "read this in an excited sales pitch voice with a slight Southern accent" and the model generates exactly that. The technical breakthrough: sub-50ms latency per request, which makes real-time conversation viable—speakers can interrupt mid-sentence and the system generates new audio in real-time without jarring delays. Early developers report Flash TTS is 40% faster than ElevenLabs Conversational v3 while offering more granular style control. Pricing: $0.50 per 1M characters (competitive with Google's existing TTS, significantly cheaper than Eleven Labs at $8 per 1M characters). Supported languages: 70+, with regional accent support (Indian English, Brazilian Portuguese, Nigerian English, etc.). Use cases exploding: podcast production (users can generate intro/outro voice variations without a voice actor), audiobook narration (style control enables distinct character voices), accessibility tools (narrate documents in the reader's preferred tone), and agentic customer service (agents use TTS to sound natural and conversational, not robotic). The market implication: ElevenLabs' competitive moat narrows. Flash TTS is enterprise-grade, cheaper, and faster. Expect ElevenLabs to respond with lower pricing or new features within 30 days. For startups building voice apps, Flash TTS is now the default choice.

ElevenLabs launched a dedicated platform for deploying conversational AI voice agents, enabling businesses to build systems that handle full customer conversations autonomously across voice, text, and multiple communication channels. The platform is built on Eleven v3 Conversational—an emotionally intelligent, context-aware text-to-speech model optimized for real-time dialogue with a new turn-taking system that produces better-timed responses and fewer interruptions. Key specs: sub-100ms latency so conversations feel like talking to a real person, support for 70+ languages, and pricing starting around 8 pence per minute. Business use cases already deployed: customer support routing, lead qualification, appointment booking. A business handling 10,000 monthly customer calls can automate 70-80% of tier-1 inquiries without live agents. ElevenLabs raised $500 million at an $11 billion valuation in February 2026, funding continued model development. International advantage: the 70+ language support with strong regional accent handling makes the platform viable for customer service operations in India, Southeast Asia, Brazil, and Germany—markets where call center costs are high and multilingual support is non-negotiable. Regulatory note: the EU AI Act's transparency requirements for automated voice systems take effect August 2026; ElevenLabs provides built-in disclosure features to help businesses stay compliant.

Google quietly launched "Google AI Edge Eloquent," an offline-first AI dictation app for iOS that processes speech entirely on-device using local Gemma AI models—no internet connection required. The app transcribes spoken words in real time, handles complex vocabulary, and can optionally import domain-specific terms (medical jargon, legal terms, industry names) from your Gmail account for improved accuracy. A cloud mode is available for enhanced accuracy using Gemini models, but users can disable it entirely for privacy or offline use. The offline capability is a breakthrough for professionals in regulated industries (healthcare, legal, finance) where cloud processing of sensitive content is prohibited. For users in regions with unreliable connectivity—rural areas, international travel, developing markets—Eloquent removes the dependency on stable broadband that has limited AI dictation adoption. Competing directly with Wispr Flow, SuperWhisper, and Willow; iOS keyboard support coming in a future update.

In a strategic pivot announced in March 2026, OpenAI shut down its Sora video creation app to focus on robotics and AI-powered world simulation. The move reflects market dynamics: Runway Gen-4, Pika 3.0, and Kling 2.5 matched or exceeded Sora's quality while cutting generation times by 60-80%. Sora's 3-8 minute generation lag for 10-second clips became untenable against competitors shipping 15-90 second turnarounds. The discontinuation signals consolidation in generative video toward platforms offering real-time iteration for content creators.

Pika released versions 2.5 and 3.0 (dual-versioning strategy to support both feature-complete and bleeding-edge users). The tool now generates sub-10-second social clips in 15-30 seconds—roughly 3-5x faster than Runway or Kling. New creative tools include Pikaffects (style transforms), Pikaframes (frame-by-frame editing), and Pikaswaps (object morphing). The free tier generates 480p video, sufficient for Instagram and TikTok iteration. Pika's positioning: rapid prototyping and social media content over cinematic polish.

Runway released Gen-4 Turbo, claiming top benchmark performance for AI video generation. The model introduces Motion Brush for frame-by-frame motion direction, advanced character consistency tracking, and industry-leading visual coherence. Pricing spans $28/month (Standard, ~625 credits) to $76/month (Pro). Runway's advantage lies in precise camera control and dynamic scene handling—essential for cinematic production. Generation speeds remain slower during peak hours, but the quality ceiling makes it the professional standard for short films and advertising.

Speak in English, your colleague hears your cloned voice in Japanese. Supports 32 languages with sub-200ms latency. Free tier includes 10 minutes per month.

ElevenLabs and competitors can now clone a voice from 30 seconds of audio. The EU proposes mandatory watermarking for synthetic speech. Several US states draft disclosure laws.