Frequently asked questions
Which voice is the most natural?
Naturalness is subjective. Aria offers a warm, clear narrator style, while Luna provides soft storytelling. All eight voices are based on licensed recordings from LibriTTS-R. You can test each voice via the /v1/voices endpoint to determine which fits your specific call center use case.
Are all voices the same price?
Yes. All eight voice presets use the same underlying engine. Whether you choose tts-1 or tts-1-hd, the price is $10 per 1 million characters. There are no extra fees for specific voices, languages, or formats like MP3 or WAV.
Can I mix male and female voices?
Yes. You can switch between any of the eight voices (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) in separate API requests. Since each request generates a distinct audio file, you can concatenate them to create dynamic IVR menus with mixed genders.
Do voices support all 23 languages?
Yes. Any of the eight voices can speak all 23 supported languages, including English, Spanish, German, and Japanese. If you do not set the language parameter, the system guesses it from the script, defaulting to English for Latin text.
Is there a voice for news updates?
Atlas is described as neutral and suitable for news. It provides a clear, professional tone ideal for delivering information. You can use the tts-1 model to generate these updates efficiently, keeping costs predictable at standard character rates.
Which voice is best for IVR menus?
Sage’s calm, instructional tone suits menu prompts well. Milo offers a friendly, conversational style for engaging interactions. Both voices support all 23 languages and are selected via the voice parameter. Since any voice can speak any supported language, you might also consider Aria or Nova if their specific timbre better fits your brand’s audio identity.
Do voices sound robotic?
Voices are based on licensed recordings from LibriVox public-domain audiobook readings, not simple concatenation. They carry an inaudible watermark marking them as synthetic. While natural, they are distinct from human speech, and you can verify the quality by testing samples.
Can I adjust the pitch of a voice?
No. The API allows you to adjust speed (0.25–4.0) but does not support pitch modification. You can also change the response format to MP3, WAV, OPUS, FLAC, or PCM. For pitch control, you would need to process the audio file after generation.
Are the voices based on real people?
Yes. The voices are based on licensed recordings from LibriTTS-R, sourced from LibriVox public-domain audiobook readings. Each generated file includes an inaudible watermark to identify it as synthetic speech, ensuring transparency about the audio's origin.
Is there a voice for children?
No. There are eight adult presets: Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion. None are specifically designed or labeled as children's voices. You can adjust the speed to alter pacing, but the vocal characteristics remain adult.
Do voices change accent by language?
Yes. Each voice maintains its character across all 23 supported languages. If you do not specify a language, the system guesses it from the script. For non-English Latin text, you should explicitly set the language parameter to ensure correct pronunciation.
Can I use a voice for ads?
Yes. Felix is energetic and great for promos, but any voice works. You can generate ad audio in any format. Since failed requests are free, you can test different voice styles without incurring costs for unsuccessful generations. Note that every file includes an inaudible watermark marking it as synthetic speech, so check your license terms before commercial use.
Is the audio watermarked?
Yes. Every generated audio file carries an inaudible watermark that marks it as synthetic speech. This helps identify the content as AI-generated without affecting the listening experience. The watermark is embedded directly into the audio response returned by the API.
Which voice has the least fatigue?
Subjective. Sage is calm and instructional, while Luna offers soft storytelling. For long IVR menus, you might prefer a neutral voice like Atlas. Test all eight voices with your specific text to determine which remains clear and engaging over time.
Criteria for Voice Selection
Choosing the right text-to-speech voice for your call center involves balancing clarity, tone, and listener comfort. Clear audio reduces misinterpretations, while an appropriate tone ensures the message aligns with your brand’s voice. Over time, the listener’s experience depends on how well these elements work together.
A voice that sounds professional yet natural helps maintain customer engagement without causing fatigue. By focusing on these core criteria, you can select a voice that supports both efficiency and a positive user experience in your IVR system.
Clarity and intelligibility
Clear speech is essential for reducing errors in customer interactions. Voices like Aria and Atlas offer distinct clarity. This clarity helps minimize repeat calls and improves overall satisfaction.
When customers can hear clearly, they spend less time repeating themselves.
- Aria offers warm, clear narration ideal for general inquiries.
- Atlas provides a neutral, news-like delivery for factual information.
- Clear audio reduces the need for customers to ask for repetition.
Tone appropriateness for business
The tone of your TTS voice should reflect your brand’s personality. Whether you need a friendly conversational tone with Milo or a formal news delivery with Atlas, aligning tone with your business identity is key.
Consistency in tone helps build trust with your customers. A mismatched tone can confuse users or make the interaction feel impersonal, which might lead to decreased satisfaction.
- Milo offers a friendly, conversational tone suitable for customer service.
- Felix provides an energetic style for promotional calls.
- Sage delivers a calm, instructional tone for guided processes.
Listener fatigue over time
Prolonged exposure to a monotonous or overly energetic voice can lead to listener fatigue. Choosing a voice that maintains a natural rhythm helps keep customers engaged throughout their interaction.
Testing different voices with real customers can reveal which options are most comfortable for extended listening. This helps in selecting a voice that supports long call durations without causing irritation.
- Nova’s bright style can be engaging but may cause fatigue if overused.
- Luna’s soft approach is ideal for storytelling or longer messages.
- Regularly review customer feedback to adjust voice selections as needed.
Female Voice Options
For call centers, voice tone directly impacts customer retention and agent efficiency. Our female presets offer distinct personalities to match your brand’s identity, from warm narration to upbeat energy, ensuring every interaction feels consistent and professional without requiring complex configuration.
Aria: Warm and clear narrator
Aria delivers a warm, clear narration ideal for general inquiries and standard IVR menus. Her balanced tone reduces listener fatigue during long hold times, making her a reliable default for high-volume automated systems.
- Best for: Standard greeting scripts, account status updates, and general information retrieval.
- Character limit: 4096 characters per request.
- Format support: MP3, WAV, Opus, FLAC, PCM.
Nova: Bright and upbeat style
Nova brings a bright, upbeat energy suitable for promotional calls or quick notifications. Her lively delivery keeps callers engaged during short interactions, such as appointment reminders or loyalty program updates, without sounding overly casual.
- Best for: Promotional announcements, appointment reminders, and short notifications.
- Speed range: 0.25 to 4.0.
- Languages: Supports all 23 available languages with consistent quality.
Luna: Soft storytelling approach
Luna offers a soft, storytelling approach that works well for detailed explanations or guided workflows. Her gentle cadence helps complex information land clearly, reducing the need for callers to ask for repetition during onboarding or support calls.
- Best for: Guided workflows, onboarding instructions, and detailed support explanations.
- Pricing: Same rate as other voices ($0.01 per 1,000 characters).
- Audio output: Returns synthetic speech with an inaudible watermark.
Male Voice Options
Male voices offer distinct tonal ranges for different call center scenarios. Felix brings high energy to promotional outbound calls, while Milo excels in friendly, conversational support interactions. Atlas provides a neutral, professional tone ideal for standard news-style updates or automated announcements.
Each voice is based on licensed recordings. They handle the 23 supported languages, maintaining their character regardless of the language spoken.
Felix: Energetic promo style
Felix is designed for high-impact messages. His energetic delivery keeps listeners engaged during short promotional bursts or urgent alerts. Use him when you need to convey enthusiasm without sounding overly theatrical.
He works well with the standard speed settings, but can handle faster paces if needed for time-sensitive announcements. His tone remains clear even at higher speeds, making him suitable for rapid-fire information delivery.
- Ideal for promotional outbound calls
- Maintains clarity at higher speeds
- High energy without being distracting
Milo: Friendly conversational tone
Milo sounds like a helpful colleague rather than a robot. He is perfect for customer support interactions where a warm, approachable tone reduces friction. His natural flow makes complex explanations feel easier to digest.
He adapts well to various languages, keeping his friendly demeanor consistent across different regions. This consistency helps build trust with callers who might otherwise feel disconnected from the service.
- Best for customer support
- Natural, approachable delivery
- Consistent tone across languages
Atlas: Neutral news delivery
Atlas delivers information with a calm, authoritative neutrality. He is ideal for standard operating procedures, account updates, or factual announcements where emotion might distract from the message. His voice is easy to listen to for extended periods.
Use Atlas for routine calls where clarity is more important than personality. He handles both short confirmations and longer informational segments with equal stability, ensuring your callers always receive clear, unbiased information.
- Neutral, authoritative tone
- Ideal for standard announcements
- Low listener fatigue
Specialized Voice Profiles
Voice selection isn't just about quality; it's about function. While general voices handle standard queries, specialized profiles optimize specific interaction types, ensuring the tone matches the intent of the call or message.
For instance, Sage provides a calm, instructional tone ideal for step-by-step guidance, while Orion offers deep, documentary-style narration for detailed updates. Matching these profiles to your brand identity creates a consistent auditory experience.
Sage for instructional content
Sage delivers a calm, instructional tone perfect for guiding users through multi-step processes. Its clarity reduces cognitive load, making it ideal for IVR menus or onboarding scripts where precise understanding is critical.
Use Sage when you need users to retain specific instructions, such as navigating a menu or verifying account details.
- Ideal for step-by-step IVR navigation
- Reduces listener fatigue during complex instructions
- Maintains clarity in noisy environments
Orion for documentary depth
Orion provides a deep, documentary-style voice suitable for delivering detailed updates or news. Its rich tone adds authority and seriousness to longer messages, such as policy changes or service outage notifications.
This voice excels when you need to convey importance and depth without sounding robotic.
- Best for detailed news or updates
- Adds authority to long-form messages
- Engages listeners with rich tone
Matching voices to brand identity
Aligning voice profiles with your brand ensures consistency across all touchpoints. Whether you choose the warmth of Aria or the energy of Felix, the voice should reflect your company's personality.
Test different profiles with real customers to see which resonates best. Document your choices to maintain consistency across campaigns, support lines, and automated calls.
- Reflect company personality
- Ensure consistency across channels
- Test with real customer feedback
Language Compatibility
Our platform supports 23 languages, from Arabic to Turkish, ensuring broad global coverage. Unlike many providers, every voice preset can speak any of these languages natively. This flexibility allows call centers to deploy consistent audio quality across diverse international markets without needing separate voice packs for each region.
Consistency is key for user experience. If you do not set the language field, the system guesses it from the script, defaulting to English for Latin text. For non-English Latin-script text, explicitly setting the language code ensures accurate pronunciation and prevents awkward misinterpretations by the engine.
23 languages supported globally
We support Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish. Any voice can speak any of these languages, giving you uniform control over tone and style regardless of the target market.
- Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish
Consistency across language switches
The same voice presets maintain their distinct character across all supported languages. This ensures that your brand voice remains recognizable whether you are serving a customer in Spanish or Japanese. Explicitly setting the language parameter in your API request guarantees the engine selects the correct phonetic rules, avoiding default English biases in Latin-script non-English text.
Accent variations in presets
Each of the eight voice presets carries a specific tonal quality that persists across languages. Aria offers a warm, clear narrator style, while Felix provides an energetic promo feel. These presets are designed to convey the right mood, whether you are delivering news with Atlas or storytelling with Luna, ensuring the accent and emotion match your message.
Note that Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali, and other languages are not currently supported. Ensure your target markets align with our 23 supported languages to avoid fallback behaviors or reduced quality.
- Aria: Warm narrator
- Nova: Bright style
- Luna: Soft storytelling
- Sage: Calm instruction
- Felix: Energetic promo
- Milo: Friendly tone
- Atlas: Neutral news
- Orion: Deep documentary
Technical Performance
Performance depends on text length and current server load. Long texts process at roughly real-time speed, averaging 25 characters per second. Short phrases may take a few seconds to generate. Since requests are processed sequentially, high traffic can introduce queuing delays.
The API uses standard audio formats for maximum compatibility. Choose from MP3, WAV, Opus, FLAC, or PCM based on your IVR system's requirements. Each request returns a direct audio file.
Processing speed per character
Generation speed scales with input length. For extended scripts, expect output roughly equivalent to real-time playback, consuming about 25 characters per second. Shorter prompts require a few seconds of processing time regardless of length.
Queue handling under load
Requests are processed one after another in a single queue. During peak usage, your synthesis job may wait for previous requests to finish. This sequential model ensures consistent resource usage but introduces variable latency under high load.
Audio format compatibility
Select the response_format parameter to match your telephony infrastructure. Supported formats include MP3, WAV, Opus, FLAC, and PCM. This flexibility allows seamless integration with various legacy and modern IVR platforms without additional conversion steps.
Pricing Impact by Voice
Your voice selection never impacts your bottom line. Every preset, from Aria to Orion, costs the same flat rate of $10 per million characters. Spaces and punctuation count as characters, but the voice type does not add a surcharge. This uniformity simplifies budgeting for call centers scaling up volume.
High-volume usage remains predictable because pricing does not fluctuate based on voice complexity. You can switch from a friendly conversational tone to a neutral news delivery without adjusting your cost-per-call calculations. This consistency supports transparent forecasting for indie developers managing tight margins.
Uniform pricing across all voices
All eight voices are priced identically. Whether you choose a warm narrator or a deep documentary voice, the charge per character remains exactly $0.01 per 1,000 characters.
Cost efficiency in high volume
Since cost is driven by character count rather than voice model, high-volume call centers benefit from consistent expenses. Failed requests are free, ensuring you only pay for successful audio generation. This protects your budget during peak traffic periods when queue delays might occur.
No premium for HD models
The 'tts-1-hd' model offers high-definition audio but costs the same as the standard 'tts-1' engine. There is no additional fee for higher quality. You can upgrade audio clarity for all your IVR prompts without increasing your per-character rate.

User Experience Considerations
User experience hinges on how naturally customers perceive the synthetic voice during interactions. While our voices are based on licensed recordings, individual preference varies significantly across demographics and call contexts. Gathering direct feedback helps identify which presets resonate best with your specific audience, ensuring higher satisfaction rates without relying on generic assumptions.
Customer preference surveys
Regularly solicit feedback on voice clarity and tone preference. Simple post-call surveys can reveal if listeners find Aria’s warmth or Milo’s conversational style more engaging for your specific use case, allowing you to tailor the experience effectively.
Accessibility and hearing aids
Ensure audio formats like WAV and FLAC are available for high-fidelity playback. Clear, uncompressed audio often works better with hearing aids and assistive devices, reducing listener fatigue during long IVR menus or detailed instructions.
Regional voice acceptance
Test voice acceptance in target regions. While any voice speaks any supported language, cultural nuances may make certain tones feel more appropriate than others. Local A/B testing helps confirm which voice feels most natural to your users.
Implementation Tips
Implementing the right voice requires a systematic approach to ensure your IVR sounds natural and consistent. By testing variations and maintaining clear scripts, you can optimize the caller experience without overcomplicating your setup.
Focus on practical adjustments that align with your existing workflow, ensuring that every interaction feels professional and recognizable to your customers.
A/B testing different voices
Use your API key to generate short clips of your greeting messages with different voices. Send these to a small group of test callers or internal staff to gather feedback on clarity and tone before rolling out changes.
Record the metrics for each test, such as drop-off rates or comprehension scores, to make data-driven decisions about which voice works best for your specific use case.
- Generate 10-second samples for each candidate voice
- Share via email or internal survey
- Track caller comprehension scores
Consistency in greeting scripts
Ensure your greeting scripts are written specifically for TTS. Avoid complex punctuation that might confuse the engine, and keep sentences concise to maintain a natural rhythm.
Use the language parameter explicitly if your text contains mixed scripts, preventing the engine from guessing incorrectly and maintaining a consistent accent throughout the call.
Stick to one primary voice for your main menu to build familiarity, reserving secondary voices for specific departments or alerts.
Updating voices seasonally
While the core voice library remains stable, consider adjusting your greeting scripts to reflect seasonal contexts. For example, a slightly warmer tone in winter or a brighter delivery in spring can enhance customer engagement.
You can swap the voice parameter in your API calls to align with these seasonal themes without needing a new integration or code overhaul, keeping your IVR fresh and relevant.
Licensing and Rights
Our voices are built on licensed recordings from the LibriTTS-R dataset (CC BY 4.0) and public-domain audiobook readings from LibriVox. We do not clone specific individuals or use celebrity likenesses, ensuring clear rights for commercial IVR and call center deployments.
The generated audio carries an inaudible watermark marking it as synthetic speech, but this does not affect your usage rights. You own the output for your specific calls without needing to track individual voice licenses per call.
LibriTTS-R source details
We license the LibriTTS-R dataset, which provides high-quality speech data under a CC BY 4.0 license. This foundation ensures that the synthetic voices we generate are built on legally sound, structured audio data suitable for commercial products.
- CC BY 4.0 licensed dataset
- High-fidelity source recordings
- Structured for TTS training
Public domain audiobook roots
Our voice models also draw from LibriVox, a collection of public-domain audiobooks. Since these recordings are in the public domain, they provide a robust and legally unrestricted foundation for creating consistent, natural-sounding synthetic voices for your call center.
- LibriVox public domain source
- No per-call attribution required
- Stable voice characteristics
Commercial use permissions
You can use the generated audio for commercial call centers, IVR systems, and customer support without additional royalty fees. The only requirement is that you hold a valid API key and adhere to our rate limits and terms of service.
- No per-call royalties
- Commercial IVR ready
- Simple API key validation
Common Mistakes to Avoid
Call centers often deploy voices that clash with the message, causing confusion or distrust. For instance, using the deep, documentary-style Orin voice for a cheerful promotional update can feel disjointed. Similarly, applying a bright, upbeat Nova voice to a serious billing error notification may minimize the perceived urgency. Always align the voice’s inherent tone with the script’s intent to ensure clarity and appropriate emotional resonance for the caller.
Mismatched tone and message
Pairing the wrong voice with specific content types can undermine your brand’s consistency. If you switch between voices without a clear strategy, callers may perceive the inconsistency as a system error rather than a stylistic choice. Define which voices handle greetings, which manage complex instructions, and which deliver promotions. Keeping voice usage predictable helps reduce cognitive load for your customers.
- Use Aria for general greetings to establish a warm, clear baseline.
- Reserve Felix for time-sensitive promotional messages where energy is key.
- Avoid using Atlas for casual, friendly updates as its neutral tone may seem too formal.
Ignoring speed settings
The API allows speed adjustments between 0.25 and 4.0, but many developers leave this at the default 1.0. For complex IVR menus, slowing the speech to 0.8 or 0.9 can significantly improve comprehension and reduce input errors. Conversely, increasing speed to 1.2 or 1.5 works well for simple confirmations. Testing these settings with real callers ensures you balance efficiency with intelligibility without causing listener fatigue.
- Set speed to 0.8–0.9 for complex menu options or instructions.
- Keep speed at 1.0 for standard informational messages.
- Increase to 1.2–1.5 for simple, non-critical confirmations to save time.
Overlooking language codes
While the API guesses language from the script, relying on this heuristic can lead to errors with Latin-script texts that are not English. For example, Dutch or German text might default to an English voice if the language field is omitted, resulting in awkward pronunciation. Always explicitly set the language parameter for non-English Latin scripts to ensure the correct voice preset applies, maintaining high quality across your 23 supported languages.
- Explicitly set the
languageparameter for Dutch, German, or Portuguese text. - Do not rely on script guessing for mixed-language messages.
- Verify voice behavior when switching between supported languages frequently.
Final Recommendations
To begin, start with Aria or Milo. Aria offers a warm, clear narrator style ideal for standard prompts, while Milo provides a friendly, conversational tone perfect for general inquiries. Both voices are versatile and handle a wide range of standard call center scripts without requiring complex configuration.
Next, test with real customer calls. Use your actual IVR scripts and route live calls through the new voices. Monitor customer feedback and drop-off rates to see how the synthetic speech performs in your specific context. This practical testing ensures the voice aligns with your users' expectations before full deployment.
Start with Aria or Milo
Aria’s warm, clear narration works well for holding music or detailed instructions. Milo’s friendly tone suits welcome messages and simple navigation prompts. Both are easy to integrate and require no special tuning to sound natural in a standard business context.
Test with real customer calls
Run A/B tests using your existing call flows. Compare hold times and customer satisfaction scores between the old TTS provider and your new API integration. Real-world usage reveals nuances that static listening tests might miss, such as fatigue or clarity issues during long waits.
Use the speed parameter (0.25–4.0) to fine-tune the delivery rate based on your test results. Adjusting speed can significantly impact how customers perceive the voice’s intelligence and patience.
- Monitor drop-off rates during voice prompts.
- Collect feedback on voice clarity and tone.
- Adjust speed settings based on user response.
Document your final choice
Once you identify the best voice for each scenario, document the configuration. Note the voice name, speed, and language settings used. This ensures consistency if your team grows or if you need to replicate the setup for other call centers.
Keep a record of the API key and endpoint details. Since you can issue a new key at any time, ensure your documentation reflects the current active key to avoid service interruptions during migration or updates.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Her gentle cadence helps complex information land clearly, reducing the need for callers to ask for repetition during onboarding or support calls.
23 languages
Any voice speaks any of these languages.
