How can I use the v3 API to pronounce single words, or very short words, in different languages? I | ElevenLabs | Page 1

AI-generated Answer for heavenly_kitten_99827

Short answer

Set language_code in your v3 TTS request to force the reading language (e.g., fr, es, ja, zh). This removes ambiguity for single‑word prompts. (elevenlabs.io)
For Japanese, optionally enable apply_language_text_normalization: true to improve character reading; note this adds latency and is currently only supported for Japanese. (elevenlabs.io)
If you need a specific per‑word reading, attach a pronunciation dictionary via pronunciation_dictionary_locators. On v3 and for non‑English words, use Alias rules (replacement spellings). SSML phoneme tags work only with English models (Flash v2, Turbo v2, English v1), not v3. (elevenlabs.io)
Pick a voice trained in your target language to get the desired accent; the accent comes from the voice, not the language_code. (help.elevenlabs.io)

Minimal v3 examples

Force French for an ambiguous single word

Model: eleven_v3
Key params: language_code: "fr"

Python (from API reference; substituting model and parameters):
client.text_to_speech.convert(
voice_id="YOUR_VOICE_ID",
text="papa",
model_id="eleven_v3",
output_format="mp3_44100_128",
language_code="fr"
)
This uses the v3 model with language forcing; v3 is supported for Text to Speech over the API. (elevenlabs.io)

Disambiguate Japanese vs Chinese readings
client.text_to_speech.convert(
voice_id="YOUR_VOICE_ID",
text="今日",
model_id="eleven_v3",
language_code="ja",
apply_language_text_normalization=True
)
Sets Japanese explicitly and enables language-specific normalization. (elevenlabs.io)
Override a single word via a pronunciation dictionary (v3/non‑English → use Alias)

Create or upload a dictionary (PLS) and get its id/version_id, then attach it per request:
client.text_to_speech.convert(
voice_id="YOUR_VOICE_ID",
text="papa",
model_id="eleven_v3",
language_code="fr",
pronunciation_dictionary_locators=[{
"id": "DICT_ID",
"version_id": "VERSION_ID"
}]
)
Attach dictionaries per request (up to 3). For non‑English or when using v3, prefer Alias rules (replacement spellings) rather than phoneme tags; phoneme SSML works only with English v1/Turbo v2/Flash v2. (elevenlabs.io)

Notes

language_code uses ISO 639‑1 (e.g., fr, es, ja, zh). If the model doesn’t support the code you pass, the API will error. (elevenlabs.io)
Avoid mixing multiple languages in the same short prompt; if you must, split requests and set language_code per call. (help.elevenlabs.io)

References

Create speech (v1/text‑to‑speech) – language_code, normalization, dictionaries. (elevenlabs.io)
Eleven v3 API usage.

#How can I use the v3 API to pronounce single words, or very short words, in different languages? I