Skip to content

API_KEY_LIMITATION_AND_SITUATION.md

Latest commit

 

History

History
99 lines (72 loc) · 5.14 KB

API_KEY_LIMITATION_AND_SITUATION.md

File metadata and controls

99 lines (72 loc) · 5.14 KB

Provider Status & Troubleshooting (April 2026)

Provider Status Summary

Provider Status Free Tier API Key Notes
Edge TTS Working Unlimited (no key) No Microsoft neural voices via edge-tts library. Always works.
gTTS (Google) Working Unlimited (no key) No Google Translate TTS endpoint. Basic quality, 100% reliable.
ElevenLabs Partial 10k credits/month Yes Free tier can use premade/default voices. Library voices blocked (402). Use eleven_flash_v2_5 model for lower cost.
Hugging Face Unavailable N/A Optional Hub lists zero TTS models on hf-inference provider as of April 2026. Every call returns 404 from router.huggingface.co. Platform limitation.
Replicate Paid only Limited free runs Yes Pay-per-use after free runs. ~$0.002-0.01/generation.
OpenAI Paid only $5 new-account credit Yes tts-1: $15/1M chars. tts-1-hd: $30/1M chars. gpt-4o-mini-tts: token-based.

Detailed Provider Notes

Edge TTS (Free, Always Works)

  • Uses Microsoft's neural TTS service via the edge-tts Python library
  • No API key, no account, no rate limits for normal use
  • 400+ voices across dozens of languages
  • Supports speed control (0.5x-2.0x)
  • Quality: very good for a free service (neural voices)
  • This is the recommended default for all users

gTTS (Google TTS) (Free, Always Works)

  • Uses Google Translate's unofficial TTS endpoint
  • No API key required
  • Supports 50+ languages
  • Quality: basic/robotic compared to neural voices
  • Rate limit: ~50-60k characters/hour before potential 429
  • Good fallback when Edge TTS is unavailable

ElevenLabs

  • Free tier: 10,000 credits/month (~10 min audio with Multilingual v2, ~20 min with Flash)
  • Voice restrictions on free tier:
    • Premade/default voices: accessible via API
    • Library/community voices: blocked (returns 402 paid_plan_required)
    • Voice cloning: requires Starter plan ($5/mo) minimum
    • Voice Design (from text description): may work on free tier
  • Models: eleven_flash_v2_5 (faster, cheaper), eleven_multilingual_v2 (higher quality)
  • Default voices expiring: December 31, 2026
  • Fix applied: Dynamic voice fetching via /api/voices/elevenlabs, flash model as default

Hugging Face (Currently Unavailable)

  • The old api-inference.huggingface.co endpoint was deprecated in late 2025
  • Replaced by router.huggingface.co/hf-inference/models/...
  • As of April 2026, the Hub filter pipeline_tag=text-to-speech&inference_provider=hf-inference returns zero models
  • This means no TTS model is served by the hf-inference provider
  • All calls return HTTP 404 from the router
  • This is a Hugging Face platform limitation, not a code bug
  • The code and fallback chain are correct; HF simply doesn't host TTS on this provider
  • Workaround: use other providers (fal-ai, replicate) through HF Hub, but those require billing

Replicate

  • No true free tier; limited free runs for testing
  • After free runs: pay-per-second GPU billing
  • Models available: StyleTTS2, OpenVoice 2, Chatterbox, MiniMax Speech
  • ~$0.002-0.01 per generation depending on model and length
  • Need to add billing at replicate.com/account/billing

OpenAI

  • $5 free credits for new accounts (no card required initially)
  • tts-1: $15/1M characters (standard quality)
  • tts-1-hd: $30/1M characters (high quality)
  • gpt-4o-mini-tts: ~$0.60/1M input tokens + $12/1M output tokens
  • 6 voices: alloy, echo, fable, onyx, nova, shimmer
  • Max 4096 characters per request

Troubleshooting

"paid_plan_required" from ElevenLabs

The voice ID you selected is a "library" voice blocked on free tier. Solutions:

  1. Use the dynamic voice list (fetched from your account)
  2. Switch to a premade/default voice
  3. Upgrade to Starter plan ($5/mo)
  4. Use Edge TTS instead (free)

404 from Hugging Face

All TTS models return 404 on hf-inference. This is not fixable from the app side. Check: https://huggingface.co/models?pipeline_tag=text-to-speech&inference_provider=hf-inference&sort=trending

402 from Replicate

You need billing enabled. Go to replicate.com/account/billing.

429 from OpenAI

Quota exceeded. Check usage at platform.openai.com/usage. Wait for reset or add credits.

Edge TTS fails

Very rare. Usually a transient network issue. Retry in a few seconds.