← Back

Changelog

New features, improvements, and fixes.

v3.0.02026-02-17
  • NewContinuous batching — parallel inference across concurrent requests
  • NewNon-streaming mode for batch TTS — improved prosody on long text
  • ImprovedFlash decode attention for faster token generation
  • ImprovedPer-slot tensor management with SlotManager for GPU efficiency
v2.8.02026-02-13
  • NewCloudflare Worker API gateway (api.murmr.dev) — auth, rate limiting, routing
  • NewOfficial Node.js SDK (@murmr/sdk) with full TypeScript support
  • NewRunPod Serverless batch endpoints for non-streaming TTS
  • NewAudio format support — MP3, Opus, AAC, FLAC, WAV, PCM via response_format
  • NewAsync jobs with webhook delivery (POST to your URL on completion)
  • NewJob status polling endpoint (GET /v1/jobs/{jobId})
  • NewLong-form audio generation with automatic chunking and retry
  • NewUsage analytics dashboard for admins
  • ImprovedInput validation and error standardization across all endpoints
v2.7.02026-02-08
  • NewRTX 5090 GPU migration with torch.compile golden cache
  • ImprovedPerformance: batched codec decoding, sequential codebooks, async codec
  • ImprovedStall detection and automatic recovery for long generations
  • ImprovedDocker golden image with pre-compiled model cache (~30s warmup)
v2.5.02026-02-02
  • NewTiered pricing system — Free, Starter, Pro, Realtime, Scale
  • NewBlog section with MDX authoring format
  • ImprovedLanding page redesign with updated value proposition
  • ImprovedSaved voice lookup optimization using database prefix queries
  • ImprovedUpdated FAQs to reflect current technical capabilities
v2.2.02025-01-29
  • NewDedicated Pod deployment with Cloudflare Tunnel (zero cold starts)
  • NewBinary WebSocket protocol — raw PCM frames (~50-100ms savings per chunk)
  • NewParallel authentication — config_ack sent immediately (~200ms savings)
  • NewAPI key caching with 5-minute TTL (~300ms auth savings)
  • ImprovedWebSocket TTFC reduced to ~460ms (server-side)
  • BreakingArchived Browser SDK (@murmr/browser-sdk) — use vanilla JS WebSocket
v2.1.02025-01-28
  • NewReal-time streaming via WebSocket
  • NewPlayground "Real-time" mode with transport selection
  • ImprovedLatency reduced from ~2.7s (SSE polling) to ~1.3s (WebSocket direct)
v2.0.02025-01-27
  • NewVoiceDesign — describe any voice in natural language
  • NewReal-time SSE streaming with ~1s time-to-first-audio
  • ImprovedMobile-responsive dashboard with slide-in navigation
  • BreakingRemoved preset voices in favor of VoiceDesign
v1.1.02025-01-26
  • NewSupport page with FAQ and contact form
  • NewAccount settings with GDPR-compliant deletion
  • NewCustom 404 and error pages
  • ImprovedAdded hreflang tags for better SEO
v1.0.02025-01-25
  • NewOpenAI-compatible API (/v1/audio/speech)
  • NewVoice Playground for testing
  • NewDashboard with API keys, usage tracking, billing
  • NewStripe billing integration
  • NewLanding page with demo in 4 languages (EN, ES, PT, DE)