I am having issues generating numbers . Everthing is fine till it’s gets to numbers then I randomly | ElevenLabs | Page 1

AI-generated Answer for ys.05

Thanks for flagging this — what you’re seeing is a known behavior when prompts contain only numbers or symbols. The language detector can misinterpret “11”, “101”, currency, dates, etc., and switch languages or output garbled speech.

What to do

If you’re using the website (no API):
- Avoid number‑only or symbol‑heavy prompts; write numbers and symbols out in words (e.g., “one hundred twenty‑three dollars”). Normalization on the website is enabled by default, but explicit words give the model clear context. (help.elevenlabs.io)
- Don’t mix languages in the same prompt; the site auto‑detects language and short/ambiguous inputs can confuse it. (help.elevenlabs.io)
- In Studio, set Text Normalization to “Always” in Project settings > Advanced. For problematic items, add rules in the Pronunciations editor (e.g., map “911” → “nine one one”). (help.elevenlabs.io)
If you’re using the API:
- Specify the language explicitly and force normalization:
  - Set language_code to your target language (e.g., "en").
  - Set apply_text_normalization to "on" (spells out numbers/dates). Example is on the Create speech endpoint. Note: on certain v2.5 models, enabling normalization has plan restrictions; see API docs. (elevenlabs.io)
- For Japanese, you can also enable apply_language_text_normalization for improved readings (higher latency). (elevenlabs.io)
- The help article explains why numbers/dates/acronyms are ambiguous and why language_code helps for short inputs. (help.elevenlabs.io)
Pick the right model/voice:
- Models differ in how naturally they read numbers. Multilingual v2 tends to handle numbers and currencies more human‑like than Flash v2.5; use it when latency allows. (elevenlabs.io)
- Use a voice trained in your target language to avoid accent drift or unintended language switching. (help.elevenlabs.io)
Clean up input:
- Avoid unusual non‑text characters (e.g., { } < > [ ]) in prompts, which can degrade speech quality. (help.elevenlabs.io)

Why this works

Numbers/symbols are ambiguous across languages; “11” can be “eleven” (EN), “once” (ES), “elf” (DE). Writing them out or fixing the language via language_code removes that ambiguity, and normalization expands numerals for clearer pronunciation. (help.elevenlabs.io)

If you can share whether you’re using the website, Studio, or API (plus model/voice and a sample input that fails

#I am having issues generating numbers . Everthing is fine till it’s gets to numbers then I randomly