ElevenLabs Text to Speech Not Working: 9 Fixes

When ElevenLabs text to speech is not working, the cause is usually one of nine things: exhausted credits, an invalid API key, text over the character limit, a model outage, or a browser issue. Work through the fixes below in order – most problems resolve in the first three steps. Plans start at $6/month for Starter with 30,000 credits (checked 5 Nov 2026).

For model-specific help, our streaming guide covers WebSocket problems, the Multilingual V2 guide covers model choice, and the WAV output guide covers format issues.

Fix 1: Check your credit balance

The most common reason generation fails is simple: no credits left. The free plan includes 10,000 credits per month, and paid plans add their allowance monthly. When the balance hits zero, the API returns a quota error and the web interface refuses to generate. Open your profile or billing page and check the remaining balance. If it is zero, wait for the monthly reset, upgrade, or buy extra credits. Our free plan limits guide explains exactly what the 10,000 credits cover.

Fix 2: Verify your API key

API calls fail with authentication errors when the key is wrong, expired or revoked. Keys start with a recognizable prefix and are shown once at creation – if you lost it, generate a new one from your profile rather than guessing. Make sure you are sending it as the xi-api-key header, not as a query parameter or in the request body. If you recently rotated keys, update every integration, script and environment variable that used the old one. A 401 Unauthorized response almost always means the key, not the text.

Fix 3: Shorten your text

Every model has a per-request character limit: 10,000 on Eleven v4, 5,000 on v3, 40,000 on Flash v2.5. Text over the limit is rejected outright. If your script is long, split it into chunks under the limit and generate each separately, then stitch the audio files. Also watch for invisible characters – text pasted from PDFs or Word can contain control characters that the API rejects. Paste into a plain-text editor first to strip formatting.

Fix 4: Try a different model

Occasionally a specific model has degraded performance or a partial outage while others work fine. If v4 fails, try v4 Turbo or Multilingual v2 with the same text. If the alternate model works, the problem is model-specific, not your account. Check the official status page or help center announcements for ongoing incidents before assuming your setup is broken.

Fix 5: Clear browser cache and try incognito

Web interface glitches – the player not loading, the generate button doing nothing, settings not saving – are often stale frontend state. Hard-refresh with Ctrl+Shift+R, clear site data for elevenlabs.io, or open an incognito window and sign in fresh. Disable ad blockers and privacy extensions for the site, since they can interfere with the audio player and API calls. If incognito works, an extension or cached file was the culprit.

Fix 6: Check your network and VPN

Corporate firewalls, VPNs and aggressive DNS filters can block the API or the WebSocket endpoints. Try disconnecting the VPN, switching networks (for example, phone hotspot versus office wifi), and confirming that api.elevenlabs.io resolves. Timeouts on long requests often indicate a proxy buffering the response – the chunked streaming endpoints are especially sensitive to middleboxes. If it works on one network but not another, the network is the problem.

Fix 7: Validate your request format

For API users, a malformed request returns a 400-level error with a message describing the problem. Common mistakes: sending text as a query parameter instead of JSON body, using a voice_id that does not exist or was deleted, passing an invalid model_id, and setting voice_settings outside the 0-1 range. Log the full error response – it usually names the exact field. Compare your payload against the API reference example for the endpoint you are calling.

Fix 8: Check voice availability

If generation fails for one voice but works for others, the voice may have been removed, made private, or restricted in your region. Cloned voices can also fail if the source audio was deleted. Try a stock voice like Rachel or Adam with the same text. If the stock voice works, recreate or reselect your custom voice. Voice library voices occasionally get delisted by their creators, which breaks saved voice_ids without warning.

Fix 9: Contact support with details

If all eight fixes fail, escalate with evidence. Note the exact error message, the endpoint or page, the model and voice used, a timestamp, and whether the problem reproduces in incognito and via the API. The help center troubleshooting section and support ticket form are the official channels. Include a minimal reproducible example – the shortest text that triggers the failure – because it lets support diagnose in minutes instead of days.

Quick diagnosis table

Pros

• Nine fixes cover the vast majority of real-world failures systematically.
• Ordered by likelihood, so most users solve it in the first three steps.
• Every fix is verifiable on official docs, status pages or error messages.
• Diagnosis table maps symptoms to causes for fast triage.
• Works for both web interface and API users.

Cons

• Cannot fix genuine service outages – those need ElevenLabs engineering.
• Some fixes require a paid plan, such as restoring a credit balance.
• Network-level issues may need your IT department, not a settings change.
• Deleted voices or keys cannot be recovered, only replaced.

Who should NOT follow this guide

Skip this checklist if ElevenLabs has announced a platform-wide outage – no amount of cache-clearing fixes a server-side incident, so check the status page first and wait. Skip it too if your issue is about billing, refunds or account access rather than generation; those go to support directly with account details. And if you are building on the API and hitting rate limits by design, the fix is upgrading your tier or slowing your requests, not debugging.

Frequently asked questions

Why is ElevenLabs text to speech not generating audio?

The most common reasons are zero credit balance, an invalid API key, or text exceeding the model character limit. Check your credits first, then the key, then shorten the text. Those three cover the majority of failures for both web and API users.

What does a 401 error mean on the ElevenLabs API?

It means authentication failed – your xi-api-key is missing, incorrect, expired or revoked. Generate a fresh key from your profile and send it as the xi-api-key header. A 401 is almost never about your text or voice settings.

Why does the generate button do nothing?

Usually stale browser state or an interfering extension. Hard-refresh, clear site data for elevenlabs.io, disable ad blockers for the site, or try an incognito window. If incognito works, an extension or cached file was blocking the request.

Can a VPN break ElevenLabs text to speech?

Yes. VPNs, corporate firewalls and proxies can block the API or interfere with streaming responses. Disconnect the VPN or switch networks to test. If generation works off-VPN, add an exception or use a different egress point.

What is the character limit per request?

It depends on the model: 10,000 characters on Eleven v4, 5,000 on v3, and 40,000 on Flash v2.5. Text over the limit is rejected. Split long scripts into chunks under the limit and generate each separately.

Why does one voice fail but others work?

The voice may have been deleted, made private, delisted from the library, or restricted. Try a stock voice with the same text. If it works, reselect or recreate your custom voice and update any saved voice_ids.

How do I report a bug to ElevenLabs?

Use the official support channels with a minimal reproducible example: the shortest text that fails, the exact error, model and voice, and a timestamp. Note whether it reproduces in incognito and via the API. Detailed reports get diagnosed far faster.

Sources checked

• elevenlabs.io/docs – help center troubleshooting section, API reference error codes and character limits, checked 5 Nov 2026.
• elevenlabs.io/pricing – plan prices and monthly credit allowances, checked 5 Nov 2026.