Frequently asked questions
Which voice is the most realistic?
Realism is subjective. Aria offers a warm, clear narrator style, while Luna provides soft storytelling. Felix delivers energetic promo tones, and Orion suits deep documentary narration. All voices are based on licensed recordings from LibriTTS-R and LibriVox. Each voice supports all 23 languages, allowing you to select the tone that best fits your chatbot’s personality.
Can I use multiple voices in one session?
Yes. You can call the POST /v1/audio/speech endpoint with different voice parameters for each text segment. Since the API returns audio files, you can programmatically concatenate these files to create a single audio stream with multiple voices within your application logic.
Do voices sound different in different languages?
Each of the eight voices supports all 23 languages, including English, Spanish, Japanese, and Arabic. The voice characteristics remain consistent, but pronunciation adapts to the language. If the language parameter is omitted, the system guesses the language, defaulting to English for Latin script text.
Is there a voice specifically for kids?
No voice is specifically designed for children. You have eight presets: Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion. While Luna offers soft storytelling, none are tailored exclusively for kid-friendly content. You must select the voice that best matches your desired tone and audience.
How do I choose between Atlas and Orion?
Atlas is a neutral male voice ideal for news and factual reporting. Orion is a deep male voice suited for documentary-style narration. Choose Atlas for clear, straight-forward information delivery. Select Orion for a more dramatic, immersive, or authoritative tone in your voice assistant applications.
Can I mix male and female voices?
Yes. You can generate audio segments using female voices like Aria or Nova, and male voices like Felix or Atlas. Since the API returns distinct audio files, you can merge these segments in your application to create conversations or mixed-narrative experiences for your users.
Are the voices based on human actors?
Yes. All eight voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and public-domain audiobook readings from LibriVox. Each generated audio file contains an inaudible watermark marking it as synthetic speech, confirming its origin from these human source recordings.
Do I need to pay extra for HD quality?
No. The models tts-1 and tts-1-hd use the same engine and price. Both are billed at $10 per 1 million characters. You can switch between standard and HD quality without any additional cost or change in your billing structure.
Can I change the voice mid-conversation?
Yes. You can change the voice for each request by specifying a different voice parameter in the JSON body. Since the API processes requests sequentially, you can switch voices between text segments, allowing for dynamic character changes or tone shifts within your application.
Is there a voice for sports commentary?
Felix is described as energetic and promo-oriented, making him suitable for sports commentary. His tone is bright and upbeat. Other voices like Nova or Milo may also work depending on the desired intensity, but Felix is specifically noted for energetic delivery.
How clear is the audio in OPUS format?
OPUS is one of the supported response formats, alongside MP3, WAV, FLAC, and PCM. It provides efficient compression. OPUS is often used for low-bandwidth scenarios.
Can I use these voices in YouTube videos?
Yes. The voices derive from LibriTTS-R (CC BY 4.0) and public-domain LibriVox readings, allowing commercial use. Each file includes an inaudible watermark identifying it as synthetic speech. You can use the audio, but you should verify specific licensing requirements for commercial use on platforms like YouTube.
Do the voices sound robotic?
No. The voices are based on licensed human recordings from LibriTTS-R and LibriVox. They are designed to sound natural and consistent. While an inaudible watermark marks them as synthetic, the audio output is intended to be clear and realistic for voice assistant interactions.
Is there a voice that sounds like a news anchor?
Atlas is a neutral male voice suitable for news. His tone is clear and professional. Other voices like Sage (calm, instructional) may also work for news-like delivery.
Female voices: Aria, Nova, Luna, Sage
Selecting the right female voice is critical for voice assistant design. Predictz Voice offers four distinct presets, each optimized for specific interaction patterns. These voices are generated from licensed LibriTTS-R recordings, ensuring consistent quality across all 23 supported languages. Each model supports speed adjustments from 0.25x to 4.0x, allowing fine-tuned control over delivery pace.
All four voices share the same underlying engine and pricing structure. Developers can swap between them without affecting cost or performance metrics. The API returns audio in multiple formats, including MP3 and WAV, suitable for various playback environments. Character limits apply per request, requiring text splitting for longer dialogues.
Aria: warm narrator tone
Aria delivers a warm, clear narrator tone ideal for guiding users through complex flows. Her vocal characteristics reduce listener fatigue during extended interactions, making her suitable for guided tours and help desks. The natural inflection patterns help maintain user engagement without sounding robotic.
Use Aria when the assistant needs to convey information clearly without demanding high energy from the user. Her voice works well in neutral tones, allowing the content to remain the focus rather than the delivery style.
- Warm, clear delivery style
- Suitable for guided tours and help desks
- Reduces listener fatigue in long interactions
- Works well in neutral tones
Nova: bright upbeat style
Nova brings a bright, upbeat energy to interactions. This voice is particularly effective for notifications, promotional messages, and quick status updates. Her energetic tone captures attention quickly, making her suitable for notifications and promos where immediate user action is desired.
Nova’s vocal profile cuts through background noise better than softer voices. This makes her ideal for mobile environments or public spaces where the assistant must compete with ambient sound. Her upbeat nature aligns well with brand voices seeking to appear approachable and active.
- Bright, energetic delivery
- Suitable for notifications and promos
- Effective in noisy environments
- Captures attention quickly
Luna: soft storytelling voice
Luna offers a soft, storytelling voice that excels in narrative contexts. She is suitable for audio content and relaxation apps where a calming presence is preferred. Her gentle pacing allows users to absorb information without feeling rushed, which is critical for bedtime stories or meditation guides.
Luna’s voice characteristics support extended listening sessions. Her soft tone minimizes auditory strain, making her a strong choice for apps focused on relaxation or detailed storytelling. She handles emotional nuance well, adding depth to scripted content.
- Soft, narrative delivery
- Suitable for audio content and relaxation apps
- Minimizes auditory strain
- Good for extended listening sessions
Sage: calm instructional voice
Sage provides a calm, instructional voice designed for clarity and precision. She is suitable for tutorials and technical support where accurate information delivery is paramount. Her steady rhythm helps users follow step-by-step instructions without confusion.
Sage’s neutral yet authoritative tone works well for educational content and technical guides. She maintains consistency across different topics, ensuring that the user’s focus remains on the instructional content rather than the voice personality. Her voice is particularly effective for complex, multi-step processes.
- Calm, precise delivery
- Suitable for tutorials and technical support
- Steady rhythm for instructions
- Effective for complex processes
Male voices: Felix, Milo, Atlas, Orion
Selecting the right male voice depends on your assistant’s specific role. Felix offers an energetic tone suitable for promotional bots and alert systems. Milo provides a friendly, conversational style ideal for customer support and companion bots. Atlas delivers a neutral, news-style delivery for factual assistants. Orion’s deep voice suits audiobooks and educational apps.
Each preset is based on licensed recordings from LibriVox public-domain audiobook readings, ensuring authentic quality without licensing surprises. All eight voices support the same 23 languages, allowing consistent voice identity across multilingual deployments. You can adjust speed between 0.25 and 4.0 to match interaction pacing.
Felix: energetic promo voice
Felix brings high energy to your voice interface, making him suitable for promotional bots and alert systems where attention is key. His tone cuts through background noise effectively, ensuring critical updates are heard. Pair him with concise text inputs under 4,096 characters for optimal clarity.
Use Felix when you need to convey urgency or excitement. His energetic delivery keeps users engaged during short interactions. Adjusting speed slightly higher can enhance the dynamic feel without distorting speech.
- Ideal for: Promotional bots, alert systems
- Tone: High energy, engaging
- Best use: Short, attention-grabbing messages
Milo: friendly conversational
Milo offers a warm, conversational style perfect for customer support and companion bots. His natural cadence makes interactions feel less robotic and more human. This voice works well for longer dialogues where user comfort is priority.
Milo’s neutral yet friendly tone reduces user fatigue during extended conversations. He adapts well to various accents within the supported 23 languages. Pair him with clear, simple text for best results.
- Ideal for: Customer support, companion bots
- Tone: Warm, approachable
- Best use: Extended dialogues
Atlas: neutral news style
Atlas provides a clear, neutral delivery suitable for news apps and factual assistants. His voice avoids emotional bias, presenting information objectively. This makes him ideal for time-sensitive updates, weather reports, or financial data.
His consistent pacing ensures listeners can absorb complex information easily. Atlas supports rapid prototyping for developers needing reliable, straightforward narration. Combine with opus or mp3 formats for efficient bandwidth usage.
- Ideal for: News apps, factual assistants
- Tone: Objective, clear
- Best use: Informational updates
Orion: deep documentary voice
Orion’s deep, resonant tone is well-suited for audiobooks and educational apps. His voice adds gravitas to long-form content, enhancing listener immersion. Use Orion for narrations where depth and authority matter.
His style works effectively with slower speeds (0.25–0.75) to emphasize detail. Orion’s clarity remains high even in noisy environments. Pair with WAV or FLAC for maximum audio fidelity.
- Ideal for: Audiobooks, educational apps
- Tone: Deep, authoritative
- Best use: Long-form narration
Voice selection by industry use case
Choosing the right voice depends on the specific interaction pattern of your voice assistant. For customer service bots, clarity and patience are critical. For narration apps, emotional range matters. For news updates, neutrality and speed are key.
Customer service bots
For customer service bots, a friendly and clear voice reduces user frustration. Milo offers a conversational tone suitable for help desks, while Sage provides a calm, instructional style for guided troubleshooting. Both ensure users understand instructions without feeling rushed.
- Use Milo for conversational, friendly interactions.
- Use Sage for clear, instructional guidance.
- Ensure high clarity for noisy environments.
Audiobook and narration apps
Audiobook apps benefit from voices with natural pacing and emotional depth. Luna’s soft storytelling voice is ideal for fiction, while Aria’s warm narrator tone suits non-fiction. These voices maintain listener engagement over long sessions, reducing voice fatigue.
- Use Luna for soft storytelling.
- Use Aria for warm narration.
- Support 23 languages for global reach.
News and weather updates
News and weather updates require neutral, authoritative voices that deliver information clearly. Atlas provides a neutral news style, ensuring facts are presented without bias. Orion’s deep, documentary voice adds gravitas to serious updates, enhancing trust in the information.
- Use Atlas for neutral news delivery.
- Use Orion for deep, authoritative updates.
- Prioritize clarity and speed for real-time updates.
Technical performance metrics
Choosing the right text-to-speech engine requires understanding its technical constraints. Generation speed averages roughly real-time, processing approximately 25 characters of text per second of audio output. Short phrases typically take a few seconds to synthesize, while longer texts maintain consistent throughput.
Latency during peak load depends on request queuing. Requests are processed one after another, so under load a request may wait in a queue. With a rate limit of 60 requests per minute per key, sustained bursts can increase latency. Audio quality remains consistent across all supported formats (mp3, wav, opus, flac, pcm), ensuring reliable playback for voice assistants regardless of the selected container.
Generation speed per character
Synthesis operates at roughly real-time speed, generating about 25 characters of text per second of audio. This predictable rate helps developers estimate latency accurately when building voice interfaces.
Latency during peak load
Requests are processed sequentially, meaning under load a request may wait in a queue. Sustained bursts exceeding the 60 requests per minute per key limit can increase latency as requests wait for processing slots.
Audio quality across formats
The API supports mp3, wav, opus, flac, and pcm formats without varying quality based on voice choice. This uniformity ensures consistent audio fidelity for voice assistants regardless of the selected container format.
Language versatility of each voice
Each of the eight voices supports all 23 listed languages without requiring separate voice models. This uniform support simplifies architecture for multilingual voice assistants, as you only need to manage one voice configuration per language rather than switching between distinct voice profiles.
The API automatically detects language from the input script if not explicitly set, though non-English Latin text defaults to English. Developers should explicitly pass the ISO code for non-English Latin inputs to ensure accurate synthesis and avoid mispronunciation errors.
All voices support 23 languages
Every voice preset, from Aria to Orion, can render text in Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, or Turkish. This allows a single voice identity to maintain brand consistency across global markets without needing to retrain or swap voice assets for each new region.
The limitation is clear: Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali, and other languages are not supported. Your assistant cannot handle these inputs with synthetic speech, so you must route them to a fallback or text display.
- 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 dialects
Since any voice can speak any supported language, dialectal consistency depends on the voice’s inherent tone rather than regional adaptation. For example, Aria’s warm narrator tone remains consistent whether speaking English or French, providing a stable auditory brand identity across different linguistic contexts.
This approach reduces complexity in your voice interface design. You do not need to maintain separate voice profiles for British vs. American English or Castilian vs. Latin American Spanish; the same voice asset handles all supported linguistic variations.
Accent handling capabilities
The voices do not offer explicit accent customization or phoneme-level control. They render the input text according to the voice’s trained pronunciation model for the specified language. If the language is not set, the system guesses from the script, which may lead to errors for mixed-language inputs.
For optimal results, always specify the language parameter in your API request, especially for non-English Latin-script text. This ensures the voice applies the correct phonetic rules and avoids defaulting to English pronunciation for words like 'café' or 'naïve'].
Pricing impact of voice choice
When building a voice interface, your budget depends on text volume, not voice personality. All eight presets—whether the warm Aria or the deep Orion—process audio at the identical rate. This uniformity simplifies cost forecasting for startups scaling their chatbot infrastructure.
Uniform pricing across voices
Every voice incurs the same cost per character. You can switch from Felix’s energetic promo tone to Milo’s conversational style without affecting your per-character rate. This consistency ensures that aesthetic choices do not introduce variable pricing tiers.
No premium for HD model
The high-definition model (tts-1-hd) uses the same engine and price point as the standard tts-1. There is no additional fee for higher fidelity audio. Developers can upgrade quality without upgrading their cost structure, keeping operational expenses predictable.
Cost per character remains same
Pricing is strictly $0.01 per 1,000 characters, regardless of voice or format. Spaces and punctuation count toward this total. Failed requests are free, ensuring you only pay for successfully generated audio, which stabilizes monthly spend for high-volume applications.
Licensing source and authenticity
Our voices originate from licensed recordings in the LibriTTS-R dataset, derived from LibriVox public-domain audiobook readings. This foundation ensures that the synthetic audio generated by our API carries a natural quality. The source material allows us to offer eight distinct voice presets.
By leveraging these established recordings, we provide a consistent auditory baseline for voice assistants without the overhead of custom voice cloning. Each voice retains the acoustic characteristics of the original speaker, ensuring reliability across diverse chatbot applications.
LibriVox public domain source
The underlying audio samples come from LibriVox, a platform hosting public-domain audiobooks read by volunteers. This ensures that the base recordings are free from restrictive copyright claims that often plague commercial voice libraries. Developers can integrate these voices with confidence, knowing the source material is widely accessible and legally clear for commercial use.
This public-domain origin means the voices are not tied to exclusive contracts with specific actors or agencies. It provides a stable, long-term solution for startups that need predictable licensing terms as their user base grows.
- Source: LibriVox public-domain audiobooks
- Dataset: LibriTTS-R
- Volunteer-read recordings
- No exclusive actor contracts
CC BY 4.0 license details
The LibriTTS-R dataset is distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. This allows commercial use provided proper attribution is given to the original contributors. Our API handles the attribution logic, so you do not need to manage complex licensing paperwork for each generated file.
You can integrate these voices into your products without paying additional royalty fees. The CC BY 4.0 terms permit modification and distribution, which aligns with our model of delivering processed audio files directly to your application.
- License: CC BY 4.0
- Commercial use allowed
- Attribution required
- No additional royalties
Watermark for synthetic detection
Every generated audio file includes an inaudible watermark that identifies the content as synthetic speech. This helps distinguish AI-generated voice output from human recordings, adding a layer of transparency for your users. It is particularly useful for compliance in contexts where AI voice usage must be disclosed.
The watermark does not affect audio quality or intelligibility. It remains transparent during playback, ensuring that the listening experience remains smooth while still providing a technical marker for synthetic content detection systems.
- Inaudible watermark
- Synthetic speech identification
- No audio quality loss
- Compliance support

Integration ease for developers
Predictz Voice simplifies integration by mirroring the standard OpenAI speech endpoint structure. Developers send a POST request to https://api.getpredictz.com/v1/audio/speech with a straightforward JSON body, requiring only the model, input text, voice preset, and response format. This familiar schema reduces onboarding time for teams already using OpenAI SDKs.
Simple JSON body structure
The API accepts standard JSON fields: model, input, voice, response_format, and optional speed and language. You specify the text payload directly, with a 4096-character limit per request. For longer scripts, split the input and join the resulting audio files. This minimal structure ensures quick parsing and reliable delivery without complex serialization logic.
OpenAI SDK compatibility
Since the request shape matches the OpenAI speech endpoint, you can use official OpenAI SDKs by simply changing the base_url to https://api.getpredictz.com/v1. Pass your tts-... API key via the Authorization: Bearer header. This allows you to reuse existing client configurations and code snippets, minimizing refactoring effort for your voice assistant infrastructure.
Rapid prototyping steps
Start by generating a short audio clip using the GET /v1/voices endpoint to list available presets. Test with a single sentence to verify latency and audio quality. Once satisfied, implement the character-splitting logic for longer dialogue trees. The consistent 60 requests-per-minute rate limit allows predictable scaling during initial testing phases without immediate need for complex queue management.
- Change base_url to https://api.getpredictz.com/v1
- Use standard Authorization: Bearer header
- Split text >4096 chars into segments
User experience considerations
Choosing the right voice directly impacts how users perceive your assistant during extended sessions. A mismatched tone can cause listener fatigue, while clarity determines usability in real-world conditions like busy streets or quiet offices.
Adjusting speed and ensuring audio quality helps maintain engagement without sacrificing comprehension or accessibility for diverse audiences.
Voice fatigue in long interactions
Extended listening sessions can lead to listener fatigue, especially if the voice lacks natural variation or feels monotonous. Selecting a voice that matches the emotional tone of your content helps maintain user interest over longer durations without causing auditory strain.
- Use Aria or Luna for narrative-heavy content to keep listeners engaged without fatigue.
- Rotate between warm and bright tones if your app supports multiple segments.
- Avoid overly energetic voices like Felix for long-form narration to prevent listener burnout.
Clarity in noisy environments
Clear articulation is critical for voice assistants used in public or noisy settings. Voices like Atlas (neutral news) or Sage (calm instructional) offer distinct enunciation that cuts through background noise better than softer tones like Luna.
Using formats like MP3 with appropriate bitrates ensures consistent audio quality across different speaker hardware.
- Prefer neutral or clear-enunciation voices for outdoor or transit use cases.
- Test audio playback on typical consumer speakers and smartphone microphones.
- Ensure volume levels are normalized to avoid sudden loudness spikes.
Pace adjustment for accessibility
Speed adjustment (0.25–4.0x) allows users to tailor the experience to their listening preferences or cognitive needs. Slower speeds aid comprehension for non-native speakers or those with hearing impairments, while faster speeds suit power users.
This flexibility is built directly into the API request, requiring no additional client-side processing.
- Set default speed to 1.0x, but allow user override.
- Use 0.75x speed for non-native English speakers or complex instructions.
- Avoid extreme speeds (<0.5x or >3.0x) unless explicitly requested by the user.
Common pitfalls in voice selection
When selecting a voice for your voice assistant, it is critical to align the tone with your brand identity. For instance, using Felix for a corporate news update may feel too promotional, while Orion might sound too heavy for a friendly chatbot.
Another common error is ignoring the language parameter for non-English Latin-script text. The API guesses English by default; failing to set the language explicitly can result in incorrect pronunciation. Additionally, overlooking the 4,096-character limit per request leads to truncated audio without proper splitting logic.
Mismatched tone for brand
Choosing the wrong voice personality can confuse users or dilute brand identity. Aria’s warm narration suits storytelling apps, whereas Milo’s conversational style works better for customer support bots. Orion’s deep documentary tone may feel inappropriate for light-hearted interactions.
- Aria: warm, clear narrator
- Milo: friendly, conversational
- Orion: deep, documentary
Ignoring language detection
The API defaults to English for Latin-script text if language is not set. For languages like Dutch or Portuguese, you must explicitly pass the ISO code. Failure to do so results in English pronunciation, degrading the user experience.
Ensure your application logic handles language detection before sending the request. This prevents awkward phrasing and ensures accurate accent handling across the 23 supported languages.
- Default language: English
- Set
languagefor non-English Latin text
Overlooking character limits
Each request supports up to 4,096 characters. Sending longer texts results in truncation unless you split the input manually. You must join the resulting audio files to create a continuous output.
Plan your text segmentation strategy to avoid cutting off sentences mid-stream. This is especially important for long-form narration or detailed instructions in voice assistants.
- Max 4,096 characters per request
- Split and join audio files
Testing strategies for voice assistants
Validating voice choices for chatbots requires systematic comparison and real-world validation. Since the API processes requests sequentially, testing under load reveals how queue delays affect user perception. You must ensure the selected voice maintains clarity and engagement during typical usage spikes without introducing excessive wait times that degrade the experience.
Collecting direct user signals helps identify subtle issues like voice fatigue or pacing mismatches. By combining controlled experiments with genuine interaction data, you can refine the audio output to suit your specific audience before full deployment.
A/B testing voice options
Run controlled experiments where different user segments hear different voices. For example, test Aria’s warm narrator tone against Felix’s energetic promo style in customer service flows. Track completion rates and drop-off points to determine which voice sustains attention longer during multi-turn conversations.
Ensure your testing framework can swap the voice parameter in the JSON body dynamically. This allows you to isolate the variable of voice timbre while keeping speed and model constant.
- Isolate voice as the single variable
- Track conversation completion rates
- Monitor drop-off points in multi-turn flows
User feedback collection
Gather qualitative data on voice clarity and fatigue. Ask users if the tone feels appropriate for the context, such as whether Luna’s soft storytelling voice suits an audiobook app. Use simple rating scales after interactions to quantify satisfaction.
Pay attention to noise environments. Verify that the chosen voice remains intelligible when background noise is present, which is critical for voice assistants used on the go.
- Rate tone appropriateness
- Measure perceived fatigue
- Check clarity in noisy settings
Performance benchmarking
Measure generation speed per character and latency during peak load. Since requests are processed one after another, queue times can increase under load. Benchmark these metrics to ensure they meet your SLA expectations for responsiveness.
Test different audio formats like MP3 or Opus to balance quality and bandwidth. Verify that the 4096-character limit works smoothly with your text splitting logic to avoid truncation errors.
- Track generation speed per character
- Measure queue latency under load
- Validate text splitting logic
Future scalability of voice choices
As your voice assistant scales, the API handles increased traffic through a straightforward queue system. Requests are processed sequentially, meaning high-volume bursts may experience slight delays during peak load, but throughput remains consistent at roughly 25 characters of text per second of generation. This predictable rhythm allows startups to architect their chatbot logic with clear expectations for user wait times.
The platform supports 23 languages with every voice capable of speaking any supported language, ensuring consistency as you expand into new markets like Spanish, Hindi, or Japanese without needing additional voice models. Since new languages are added to the core API rather than requiring new voice assets, scaling geographically is seamless and does not complicate your integration code.
Handling increased traffic
The API processes requests one after another, so under heavy load a request may wait in a queue. This linear processing ensures stability but means latency increases with concurrency. Startups should implement retry logic with exponential backoff to manage these queues effectively during traffic spikes without overwhelming the service.
- Sequential processing ensures stable performance
- Typical speed: ~25 chars/sec of generation
- Implement retry logic for queue delays
Adding new languages
All eight voices support the full list of 23 languages, so adding a new language to your app does not require integrating new voice models. You simply pass the appropriate ISO language code in the request. This uniformity simplifies your codebase as you expand into markets like Turkish, Korean, or Swahili.
- 8 voices cover all 23 languages
- No new models needed for new languages
- Use ISO codes to specify language
Updating voice catalogs
Voice catalogs are updated centrally by the API provider, not per client. You do not need to manage individual voice assets or update your SDKs when new presets are added. Your integration remains stable as the provider handles backend updates, allowing you to focus on your assistant's core logic.
- Centralized voice updates
- No client-side SDK updates required
- Stable integration during catalog changes
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. This allows a single voice identity to maintain brand consistency across global markets without needing to retrain or swap voice assets for each new region.
23 languages
Any voice speaks any of these languages.
