Skip to main content
Multilingual and Hinglish calls in DialNexa work best when language, transcriber, voice, prompt examples, and test data are chosen together. For Hindi-English calls, compare Soniox and Deepgram Flux multilingual on real recordings, then use Hinglish Map when the caller base naturally switches between Hindi and English. DialNexa agent builder showing a Hindi-English voice setup with GPT-4.1 and Soniox selected. DialNexa Speech Settings panel showing the Hinglish Map button below Denoising Mode. DialNexa Hinglish Map modal showing custom dictionary, global dictionary entries, search, and add word controls.
Mixed-language callers do not wait for your architecture diagram. They switch languages because that is how they speak. The agent should keep up.

​
DialNexa Multilingual And Hinglish Call Stack

​
Language Selection Details

The dashboard sorts English first, Hindi second, Indian languages next, and then other languages alphabetically. That ordering keeps common India-focused Voice AI setups close to the top without hiding other supported languages.

​
Short Replies Keep The Conversation Language

Short replies such as “good morning” or “no, thank you” do not by themselves force an established Hindi conversation into English. Turns of up to three syllables carry forward the previous language decision. Test a complete sentence when checking an intended language switch, and make sure the new language is configured for the agent.

​
Soniox Versus Deepgram For Multilingual Calls

Do not choose a transcriber from an English demo alone. Fast and wrong is still wrong.
For Hindi-English agents, DialNexa sends Soniox the valid language hints en and hi. It does not send combined hint codes such as hi-en. For supported single-language Soniox calls, DialNexa sends the selected language code together with English. For broad or unsupported language choices, it avoids strict hints instead of passing an invalid code. Deepgram Flux multilingual uses repeated language hints rather than one combined language value. When the selected language is Hindi-English, DialNexa derives hi and en hints. When a language is unsupported or intentionally broad, Flux can run without strict hints and let the provider auto-detect among supported languages.

​
Requests To Switch Language

Configure the primary language and up to three secondary languages before testing language changes. With a supported configured language, the agent can follow an explicit request such as “Please speak Hindi” and keep that reply language until the caller asks otherwise. A request does not add a new language to the configured voice or transcriber. If the caller uses an unsupported language, the agent falls back to its primary language and can explain which languages it supports. Test this with your real prompt, voice, and caller phrases rather than assuming every language can be detected correctly. Questions about the agent’s identity or supported languages are allowed even when unrelated topics are restricted.

​
Test Numbers In Multilingual Replies

Number speech now uses language-aware conversion for Hindi, Kannada, Marathi, Gujarati, Bengali, Telugu, and Tamil, with Punjabi using the Hindi number tables. Test dates, times, currency amounts, phone numbers, and one-time codes separately. Phone numbers and codes should be spoken digit by digit; preserve identifiers and email addresses rather than rewriting them into ordinary number words. For Soniox, English is included alongside configured language hints to help with mixed vocabulary. Language hints improve recognition but do not expand the languages your voice is configured to speak.

​
Hinglish Map

Hinglish Map is for phrasing, not translation of the whole agent. Use it to replace formal Hindi terms with natural mixed-language words that your callers actually use. Language tracking runs independently of whether you added Hinglish Map entries. DialNexa uses the selected languages and available transcript-language evidence to guide each reply. For Hindi-English configurations, this includes English, Devanagari Hindi, Romanized Hindi, and mixed Hinglish. Formal-Hindi substitutions apply when Hindi or Hinglish is selected and the reply is Hindi or Hinglish. They do not rewrite an English reply or activate simply because a dictionary exists. The language sent to a compatible speech provider follows the reply language, which may differ from the caller’s language when your agent cannot support the requested language.
Language tracking helps the agent choose the response style. It does not change the selected TTS voice. Use a voice that can speak the language and script you expect callers to use.

​
Write Prompts For Mixed Language

1

Name the expected language behavior

Tell the agent whether it should start in Hinglish, mirror the caller, or stay mostly English with Hindi support.
2

Include sample caller phrases

Add phrases that callers actually say, including short replies, local terms, city names, and English product words.
3

Include sample agent replies

Show the tone you want. One good example can prevent twenty awkward formal sentences.
4

Protect important terms

Keep legal, medical, financial, or compliance terms precise. Natural language is good. Sloppy meaning is not.

​
Review Checklist

​
Common Multilingual Mistakes

The model may answer in English even when the caller expects Hinglish. Add explicit language instructions and examples.
Use Hinglish Map and test another voice or model. The selected speaker matters as much as the text.
A city, plan, date, or payment amount injected through variables may need wording around it so it sounds natural.
Field descriptions should describe the business meaning, not the exact language used during the call.

Supported Transcribers

Compare Soniox and Deepgram options.

Speech Settings

Use Hinglish Map and Response Eagerness.

Text To Speech

Choose a voice that fits mixed-language calls.

Transcripts And Recordings

Audit mixed-language evidence.