Frequently asked questions
What is Hindi text-to-speech?
NavVoce’s Hindi TTS converts text into natural-sounding audio using licensed recordings from LibriVox. It supports Hindi alongside 22 other languages, allowing product teams to generate clear navigation prompts. The service delivers synthetic speech suitable for in-car listening environments without requiring custom voice cloning or SSML.
How is Hindi detected in the API?
The API guesses the language from the script if the language field is omitted. However, for Hindi text using Latin script, you must explicitly set the language parameter to ensure correct pronunciation. This ensures accurate delivery of navigation instructions regardless of the input format.
Which voice sounds most natural for navigation?
For navigation, Sage offers a calm, instructional tone, while Atlas provides a neutral, news-style delivery. You can test all eight presets, including female voices like Aria and Nova, to find the best fit for your infotainment system’s user experience.
Is the latency low enough for driving?
Typical speed is roughly real-time, generating about 25 characters of text per second. Short phrases take a few seconds, while longer texts process sequentially. Under load, requests may queue, but the output remains synchronized with driving speeds, ensuring timely navigation updates without significant delay.
Can I use MP3 for Hindi navigation prompts?
Yes, MP3 is one of the supported response formats alongside WAV, OPUS, FLAC, and PCM. You can specify mp3 in your API request to receive compressed audio files suitable for efficient storage and playback in automotive infotainment systems, ensuring broad compatibility with existing hardware.
How do I handle mixed Hindi-English text?
Set the language parameter to the primary language of your text to ensure accurate pronunciation. The API supports 23 languages, including Hindi and English. Explicitly defining the language prevents the system from defaulting to English for Latin-script text, ensuring mixed-language navigation prompts sound natural and correct.
What is the character limit for Hindi text?
Each API request supports up to 4,096 characters, including spaces and punctuation. For longer navigation instructions, split the text into parts of this size and join the resulting audio files. This allows you to generate detailed, multi-step directions without exceeding the single-request character limit.
Does the API support SSML for Hindi?
No, the NavVoce API does not support SSML. It accepts plain text input via the input field. You can control speed using the speed parameter (0.25–4.0) and select from eight voice presets, but advanced markup like pause tags or pronunciation rules is not available.
Can I clone a specific Hindi voice?
No, custom voice cloning is not offered. NavVoce provides eight preset voices based on licensed recordings. If you need a specific tone, you can choose from existing options like Sage or Atlas. Voice cloning requires uploading your own voice, which is not a feature of this API.
How much does Hindi TTS cost per request?
Pricing is $10 per 1 million characters, or $0.01 per 1,000 characters. Spaces and punctuation count toward the total. Failed requests are free, with charges returned in full. You can start with a $0.50 free trial credit valid for seven days upon Google sign-in.
What happens if the API fails during generation?
If synthesis fails, the charge is returned in full. You are not billed for unsuccessful requests. This ensures you only pay for successfully generated audio files. You can retry the request with the same parameters or adjust the input text to resolve any generation issues.
Is the audio stored after generation?
No, generated audio is returned in the response and not stored by NavVoce. Input text is used only to produce the audio and is not used for training. Request metadata like time and status is kept for billing purposes, but the audio files themselves are transient.
Do I need an internet connection for TTS?
Yes, the API is cloud-based. You must send a POST request to the endpoint to receive audio. There is no offline mode or local SDK. An internet connection is required for every synthesis request, as processing happens on NavVoce’s servers before the audio file is returned.
Can I use this API for offline navigation?
No, the API requires an internet connection for each request. Since audio is generated in the cloud and returned as a file, it cannot be used for purely offline navigation without caching. You would need to pre-generate audio files while online and store them locally in your app.
Language Detection and Specification
NavVoce’s API detects the script of your input text to guess the language, defaulting to English for Latin characters. For Hindi, which uses the Devanagari script, the system recognizes the characters automatically, so you do not need to set the language field.
Automatic script detection
When you send text in Devanagari, the API identifies the script and applies the appropriate phonetic rules. This works seamlessly for pure Hindi inputs, allowing your navigation prompts to render correctly without needing extra configuration steps for standard language inputs.
Explicit language code usage
For mixed-language routes or when using Latin-script transliterations of Hindi, explicitly set the language field to hi in the request body. This prevents the system from guessing incorrectly, ensuring consistent pronunciation even when English words are mixed into the Hindi prompt.
- Set
language: "hi"for Devanagari or mixed inputs. - Prevents default English fallback for Latin-script Hindi.
- Ensures consistent voice engine selection.
Accuracy in mixed languages
Hindi navigation often mixes English road names or landmarks. The API handles this by respecting the explicit language code, though complex mixed-script inputs may benefit from explicit tagging. For best results, keep route instructions concise and avoid excessive code-switching within a single 4096-character request.
Voice Selection for Hindi Prompts
NavVoce provides eight voice presets that speak Hindi, ensuring your navigation prompts feel natural to Indian drivers. Each voice offers a unique tonal character, allowing product teams to select a style that matches their brand’s personality without sacrificing intelligibility.
Available Hindi voices
You can choose from Aria, Nova, Luna, and Sage for female voices, or Felix, Milo, Atlas, and Orion for male voices. All eight presets support Hindi natively, so you do not need a separate Hindi-specific voice model. This uniformity simplifies your engineering workflow and ensures consistent audio quality across all languages.
- Aria: Warm, clear narrator suitable for detailed directions.
- Nova: Bright and upbeat for quick alerts.
- Luna: Soft and storytelling for longer route summaries.
- Sage: Calm and instructional for complex maneuvers.
- Felix: Energetic and promo-style for brand announcements.
- Milo: Friendly and conversational for standard navigation.
Tone and clarity differences
In a noisy cabin, clarity is critical. Sage’s calm tone reduces listener fatigue during long drives, while Felix’s energetic delivery captures attention for sudden alerts. Nova’s brightness works well for short, high-priority notifications like "Turn right in 200 meters." Atlas offers a neutral, news-like delivery that many fleet managers prefer for consistency.
Choosing between female and male voices
The choice depends on your target demographic and brand identity. Male voices like Orion provide a deep, authoritative tone often associated with reliability, while female voices like Aria offer a warm, approachable feel. You can test both in your app to see which reduces cognitive load for your users. Remember, every voice can speak all 23 supported languages, so you can mix voices if your app supports multilingual regions.
Latency in In-Car Environments
In-car navigation demands clear, timely audio cues without distracting the driver. NavVoce’s TTS engine generates audio at roughly real-time speed on long texts, producing about 25 characters of text per second of audio. This natural pacing ensures that turn-by-turn instructions sound like a companion rather than a robotic list, maintaining driver focus during complex maneuvers.
Real-time generation speed
For typical navigation prompts, generation is nearly instantaneous. Short phrases may take a few seconds to synthesize, but complex route details are processed efficiently. This speed aligns with human conversation rates, making the synthetic voice feel integrated into the driving experience rather than an external interruption.
Impact of network latency
Since the API is cloud-based, network conditions influence the total time from text to audio playback. In areas with strong cellular coverage, requests are processed quickly. However, in tunnels or rural zones, latency may increase slightly, potentially delaying the start of audio playback. Product teams should buffer audio data to mitigate this.
- Strong signal: minimal delay.
- Weak signal: potential buffering needed.
- Cloud dependency requires stable connectivity.
Handling queue delays
Requests are processed sequentially. Under heavy load, a request may wait in a queue before synthesis begins. For navigation, this means slight pauses during peak usage times. To maintain smooth audio flow, design your application to pre-fetch or cache common navigation phrases, reducing reliance on real-time API calls for every instruction.
- Sequential processing model.
- Queue waits possible under load.
- Pre-fetching mitigates delay.
Audio Format Optimization
Choosing the right audio format is critical for in-car navigation systems where bandwidth and clarity must balance. NavVoce supports MP3, WAV, OPUS, FLAC, and PCM, allowing your team to select the optimal stream for your vehicle’s infotainment architecture.
MP3 vs WAV for dashboards
WAV provides uncompressed, pristine audio ideal for high-end dashboards with ample bandwidth. MP3 offers smaller file sizes and faster transmission, making it suitable for standard infotainment units where network conditions vary.
- WAV: Suitable for local playback and high-fidelity systems.
- MP3: Suitable for streaming over cellular connections.
Bandwidth considerations
In-vehicle networks often share bandwidth with other systems. Selecting OPUS or MP3 reduces data usage, ensuring navigation prompts play without lagging behind the vehicle’s current position, even in areas with weaker signal.
- Use OPUS for efficient compression.
- Use FLAC for lossless quality when bandwidth permits.
Quality vs size trade-offs
PCM offers raw data with no compression but results in larger files. FLAC provides lossless compression for smaller sizes. Choose based on your target vehicle’s storage capacity and network reliability.
- PCM: Raw, high-size files.
- FLAC: Lossless, compressed size.
Text Processing for Navigation
Navigation prompts often mix numbers, addresses, and proper nouns. Our API processes standard Latin and Devanagari scripts directly, but explicit language specification ensures consistent pronunciation. For mixed-language inputs like 'Turn left on NH-48', setting the language parameter to hi prevents English defaults and maintains clarity for Hindi-speaking drivers.
Long route instructions exceed the 4096-character limit per request. Your backend must split these into smaller chunks before sending them to the API. The generated audio files are then joined sequentially to create a seamless instruction. This approach keeps latency low while ensuring no part of the direction is truncated.
Handling numbers and addresses
The API reads numbers as they appear. For example, '5 km' is spoken naturally in Hindi. However, ambiguous addresses benefit from explicit language codes to avoid misinterpretation of proper nouns.
Mixed-language strings are common in Indian navigation. Using the language field ensures the engine prioritizes Hindi phonemes over default English rules, reducing awkward pauses or incorrect stress patterns.
- Use
language: 'hi'for mixed inputs - No SSML tags required
- Numbers are read naturally
Splitting long route instructions
Since each request supports up to 4096 characters, complex multi-step directions must be divided. Your application should handle the concatenation of the resulting audio files on the client or server side.
This splitting strategy ensures that even lengthy routes with many turns are processed efficiently. It also allows for better error handling if one segment fails to generate.
- Max 4096 characters per request
- Join audio files sequentially
- Handle splitting in your backend
Pronunciation clarity
Voices like nova or aria offer clear enunciation suitable for navigation. atlas provides a neutral tone for straightforward instructions. Choose voices that minimize ambiguity in high-noise car environments.
Test your prompts with native Hindi speakers to ensure natural flow. The API’s synthetic nature means it does not infer context beyond the text provided, so precise wording is key.
- Test with native speakers
- Choose voices for clarity
- Precise wording reduces errors
Technical Integration Steps
Integrating NavVoce’s Hindi TTS into automotive navigation systems requires precise API configuration and structured request handling. This section outlines the technical steps for connecting your dashboard to our speech endpoint, ensuring seamless delivery of clear, natural-sounding Hindi instructions for drivers.
API endpoint configuration
Point your HTTP client to the base URL https://api.cph2699.info/v1. Include your unique API key in the Authorization header as Bearer tts-.... This key authenticates your account and tracks character usage against your balance. Ensure your application handles standard HTTP status codes to detect authentication or rate-limit errors early.
Request body structure
Send a POST request with a JSON body containing model, input, voice, and response_format. For Hindi navigation prompts, explicitly set the language field to hi to ensure accurate pronunciation, as automatic detection may default to English for Latin-script numbers. You can adjust the speed parameter between 0.25 and 4.0 to fit dashboard timing constraints.
Select a voice like sage for clear instructions or felix for energetic prompts. The input field accepts up to 4096 characters. For longer route directions, split the text into smaller chunks, generate each audio file, and concatenate them in your application logic before playback.
- model: "tts-1" or "tts-1-hd"
- voice: "sage", "aria", "nova", etc.
- language: "hi" for Hindi
- response_format: "mp3" or "wav"
Response parsing

Testing Hindi TTS in Vehicles
Testing Hindi TTS in vehicles requires validating audio clarity under real-world noise and latency conditions. Product teams must ensure that navigation prompts remain intelligible during turns, highway merges, and cabin conversations.
We help teams simulate these environments using controlled playback tests and live driver trials to verify that voice selection and language codes perform as expected in diverse driving scenarios.
Simulating in-car conditions
Use noise-cancelling headsets or cabin microphones to replicate highway and city driving sounds. Generate Hindi prompts via the API and play them through the vehicle’s audio system at varying volumes to test clarity against engine and wind noise.
Validate that the language field correctly identifies Hindi script and that the chosen voice (e.g., Luna for storytelling or Atlas for news-style instructions) maintains a natural tone without distortion.
- Test at 50–70 dB ambient noise levels
- Verify volume normalization across devices
- Check for audio clipping during sharp turns
User testing with drivers
Conduct live drives with actual drivers to assess cognitive load and comprehension. Ask participants to repeat instructions after hearing them to gauge accuracy. Ensure that Hindi prompts are not confused with English or other languages.
Use the API’s speed parameter (0.25–4.0) to find the optimal balance between clarity and pace. Typical generation speed is real-time, but queue delays under load may affect responsiveness.
- Measure reaction time to prompts
- Assess voice preference among drivers
- Record user feedback on tone
Feedback loops
Collect driver feedback on voice clarity, speed, and naturalness. Use this data to adjust voice selection or speed settings. Integrate feedback into your development cycle to refine Hindi TTS integration.
Monitor API usage metrics (characters per request, failure rates) to identify any technical issues. Ensure that audio files are not stored beyond necessary for billing purposes, maintaining user privacy.
- Track API request success rates
- Log user-reported mispronunciations
- Iterate on voice and speed settings
Common Issues and Solutions
Even with accurate language detection, edge cases in Hindi address formats or mixed English-Hindi prompts can cause slight mispronunciations. Since the API does not support custom pronunciation lexicons, developers must rely on standard text formatting to ensure clarity.
Mispronunciation handling
Text normalization helps reduce errors. For complex addresses, spelling out numbers or using standard abbreviations improves synthetic speech accuracy. If a specific phrase sounds unnatural, rephrasing the input text is the most reliable workaround until the next model update.
- Spell out complex numbers explicitly
- Use standard abbreviations for street names
- Rephrase ambiguous mixed-language prompts
Voice mismatch errors
The system guesses the language from the script, but Latin-script Hindi text defaults to English pronunciation. Always set the language field to hi when using Latin script for Hindi. This ensures the correct voice model applies to the input text.
- Always set
language: hi - Verify voice selection matches intent
- Check script detection for Latin text
Audio clipping
Clipping occurs when audio levels exceed the dashboard's maximum input threshold. Use the response_format parameter to select wav for uncompressed quality or mp3 for smaller files. Adjusting the speed parameter can also help fit longer instructions into the available audio buffer without distortion.
- Use
wavfor uncompressed quality - Adjust
speedto fit buffers - Test with dashboard max input levels
Cost Implications for Hindi TTS
Hindi TTS billing is straightforward: you pay $10 per 1 million characters, which equals $0.01 per 1,000 characters. Every character counts, including spaces and punctuation, so precise route instruction formatting directly impacts your monthly spend.
Character count impact
Each navigation prompt is charged based on its exact length. Short, concise instructions are more cost-effective than verbose descriptions. Failed requests are free, so retry logic does not incur extra charges.
Speed settings and cost
Adjusting the speed parameter (0.25–4.0) changes playback duration but does not affect character count or cost. Whether you speed up a Hindi turn-by-turn instruction or slow it down for clarity, the price per character remains constant.
Volume discounts
Our pricing is flat-rate with no tiered discounts based on volume. You pay the same rate regardless of whether you generate 1,000 or 10 million characters. There are no monthly plans or minimum spend requirements.
Data Privacy in Navigation Apps
When you send navigation instructions to our API, the text is processed solely to generate the audio file. We do not use your input strings for training our models or improving the voice engine, ensuring that specific route data remains private to your application's immediate use.
Input text usage
Your text payload is transient. Once the audio is synthesized and returned in the response, the original text is not retained for long-term storage. This ensures that sensitive address data or user preferences spoken aloud are not archived in our training datasets.
Audio storage policies
The generated audio file is delivered directly in the API response and is not stored on our servers. We only retain request metadata—such as timestamp, character count, and status—for billing purposes. The actual audio content is ephemeral, reducing your data footprint.
User data protection
Since we do not store the audio or the input text, there is no risk of a data breach exposing your navigation logs. We keep minimal metadata for billing transparency, which includes the time of the request and the volume used, but nothing identifiable.
Future Updates for Hindi Support
We are continuously refining the Hindi engine to better handle the complex phonetic structures found in navigation prompts. Our focus is on delivering clearer, more natural-sounding directions that integrate seamlessly into the in-car environment without requiring constant user adjustment.
New voice additions
We are evaluating additional voice presets optimized for Hindi speech patterns. These new voices will offer distinct tonal qualities, giving product teams more flexibility to match their specific brand identity within the vehicle interface.
- Expanding the library beyond the current 8 presets
- Focusing on voices that reduce listener fatigue during long drives
Improved accuracy
Accuracy improvements target the nuanced transitions between Hindi and English code-switching, which is common in urban navigation. We are enhancing the model’s ability to correctly interpret mixed-script inputs, ensuring addresses and street names are pronounced correctly without manual tagging.
- Better handling of mixed Hindi-English inputs
- Reduced mispronunciation of complex address formats
Format expansions
While we currently support MP3, WAV, Opus, FLAC, and PCM, future updates may introduce new compressed formats better suited for low-bandwidth in-car connections. This ensures high-quality audio delivery even in areas with weaker cellular signals.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. A neutral tone works best for factual updates, while warmer tones suit conversational interactions.
23 languages
Any voice speaks any of these languages.
