Hi, I’m using the /v1/text-to-speech/{voice_id}/stream endpoint. About 50 % of my API calls return

[hollow swift]

AI-generated Answer for leolinglian

Thanks for the details. A 401 on POST /v1/text-to-speech/{voice_id}/stream is almost always an auth/entitlement issue. Based on ElevenLabs’ docs, check the following:

• xi-api-key header on every request: The streaming endpoint requires the xi-api-key header. If it’s missing/typo’d on any call, you’ll get 401 invalid_api_key. Verify the header is set per request (not just per client). (elevenlabs.io)
• API key restrictions/permissions: Keys can be restricted by endpoint (“scope restriction”). Ensure the key you’re using is allowed to call Text to Speech (text_to_speech permission if you’re using service-account keys). Otherwise requests will fail. (elevenlabs.io)
• Quota/credit limits: Exceeding quota can be surfaced as 401 Unauthorized. Confirm your workspace and/or key-level limits (character_limit on service-account keys) and consider enabling usage-based billing if needed. (elevenlabs.io)
• Voice ID validity: Using an invalid or inaccessible voice_id can present as a 400/401 per the help article; double-check the voice exists and is accessible to the key/workspace. (help.elevenlabs.io)

Quick isolation steps:

• Call GET /v1/models with the same client and headers. If this intermittently 401s, it’s an auth/entitlement/header propagation issue rather than voice/model input. (elevenlabs.io)
• If you’re testing load, note that concurrency issues return 429 (too_many_concurrent_requests or system_busy), not 401—so a 401 points back to the items above. (help.elevenlabs.io)

If you share how you set the header (client/library, where the key is injected) I can suggest a minimal repro to catch any per-request header drops.

Sources:
Stream speech | ElevenLabs Documentation
Authentication | ElevenLabs Documentation
May 5, 2025 | ElevenLabs Documentation
API - Error Code 400 or 401 – ElevenLabs
Authentication | ElevenLabs Documentation
API - Error Code 429 – ElevenLabs

[oblique salmon]

public byte[] generateAudioBytes(String text, String voiceId) {
try {
String actualVoiceId = voiceId != null ? voiceId : defaultVoiceId;
if (actualVoiceId == null) throw new AiChatException("VoiceId is required");

        JSONObject body = new JSONObject();
        body.put("text", text);
        body.put("model_id", defaultModelId);

        String url = String.format("%s/v1/text-to-speech/%s/stream", apiHost, actualVoiceId);
        HttpHeaders headers = new HttpHeaders();
        headers.setContentType(MediaType.APPLICATION_JSON);
        headers.set("xi-api-key", apiKey);

        log.warn("Request url: {}, Send body: {}", url, body.toJSONString());
        HttpEntity<JSONObject> entity = new HttpEntity<>(body, headers);

        ResponseEntity<byte[]> response = sslRestTemplate.exchange(
                url,
                HttpMethod.POST,
                entity,
                byte[].class
        );

        if (response.getStatusCode() == HttpStatus.OK) {
            return response.getBody();
        } else {
            throw new AiChatException("TTS API failed with status: " + response.getStatusCode());
        }

    } catch (Exception e) {
        log.error("Failed to process ElevenLabs TTS", e);
        throw new AiChatException("Error generating ElevenLabs audio");
    }
}