Frequently asked questions
What are the 8 voices?
The API offers eight presets: Aria, Nova, Luna, and Sage for female voices; Felix, Milo, Atlas, and Orion for male voices. Each voice has a distinct tone, from warm narration to deep documentary styles, allowing you to select a fit for your manuscript’s mood.
Which voice is best for romance?
There is no objectively best voice for romance. Aria offers a warm, clear narrator style, while Luna provides a soft, storytelling tone suitable for romance. You can adjust the speed between 0.25 and 4.0 to enhance the delivery. The voices are based on licensed recordings from LibriTTS-R and LibriVox, ensuring a consistent, natural sound quality for your audiobook chapters.
Can I mix voices in one book?
Yes, you can mix voices by sending separate requests for different sections of your text. Since each request has a 4096-character limit, you split longer texts into parts. You then join the resulting audio files to create a single audiobook with varied narrator voices.
What is the difference between Aria and Luna?
Aria is described as a warm, clear narrator, ideal for straightforward reading. Luna is soft and storytelling-focused, better for emotional or narrative-heavy content. Both are female voices but serve different tonal needs within your audiobook production workflow.
Is Milo good for non-fiction?
Milo is a friendly, conversational male voice, making him suitable for approachable non-fiction. For a more neutral, news-style tone, Atlas is available. For deep, documentary-style narration, Orion is an option. Choose based on whether you prefer conversational or authoritative non-fiction delivery.
Do voices sound robotic?
Voices are based on licensed LibriTTS-R recordings from LibriVox, offering clear, natural-sounding speech ideal for audiobooks. Every generated file includes an inaudible watermark to mark it as synthetic speech, ensuring transparency without compromising the listening experience. While we do not claim they are entirely non-robotic, the audio is designed to be engaging and easy to listen to.
Can I change the voice mid-sentence?
No, you cannot change the voice mid-sentence within a single request. You must split your text into separate requests, each specifying a different voice. Then, you join the resulting audio files. This allows voice changes between sections or paragraphs, but not within a single sentence.
What is the speed range?
You can adjust the speech speed between 0.25 and 4.0. This allows you to slow down complex passages or speed up lighter content. The default speed is suitable for most audiobooks, but fine-tuning helps match the pacing to your manuscript’s tone and audience preferences.
Are there age variations?
The API does not offer specific age variations like child or elderly voices. You have eight presets with distinct tones, such as deep or bright. You can adjust speed and language to influence perception, but the core voice models remain fixed to their defined characteristics.
Do voices speak all languages?
Yes, every voice can speak any of the 23 supported languages, including English, Spanish, French, German, Japanese, and Arabic. If you do not set the language, it is guessed from the script. For non-English Latin text, explicitly set the language code for accuracy.
What is LibriTTS-R?
LibriTTS-R is a dataset of licensed recordings from LibriVox public-domain audiobook readings, used under CC BY 4.0. Our eight voice presets are based on these recordings. This foundation distinguishes our synthetic speech from purely algorithmic generation, providing clear audio for your audiobooks without requiring complex technical setups. Generated files carry an inaudible watermark marking them as synthetic speech.
Is the voice cloning available?
Custom voice cloning is not available. You cannot upload your own voice or design a custom voice. You must choose from the eight preset voices (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) provided by the API for your audiobook production needs.
Can I hear samples?
The API returns audio files in formats like MP3, WAV, OPUS, FLAC, or PCM. You can generate short samples by sending small text inputs to the API endpoint. This allows you to test voices and speeds before committing to full chapter generation.
How do I choose a voice?
Select a voice based on your manuscript’s tone. Use Aria for warm narration, Luna for soft storytelling, or Atlas for neutral news. Consider the audience and genre. Generate samples with different voices to hear how they fit your specific text before proceeding with full production.
Do voices have accents?
Any voice can speak any of the 23 supported languages, so dedicated accents are not tied to specific voices. For example, the English voice Aria can naturally speak French or German. While we do not offer specific accent controls, you can adjust the speed to influence the delivery style.
What format is best for quality?
FLAC and WAV offer lossless quality, while MP3 is compressed for smaller files. PCM provides raw audio. All formats are supported. Choose FLAC for archival preservation, WAV for editing compatibility, or MP3 for efficient storage. Quality depends on your specific distribution needs rather than a single superior option.
Is there a demo?
You receive a $0.50 free trial credit (about 50,000 characters) valid for 7 days when signing in with Google. Use this credit to generate audio samples with different voices and formats. This allows you to test the API’s quality before purchasing additional credits.
Can I use multiple voices?
Yes, you can use multiple voices by sending separate API requests for each voice segment. Split your text into parts under 4096 characters, assign a voice to each part, and then join the resulting audio files. This enables dynamic narration with varied voices throughout your book.
How do I test voices?
Send a short text input to the API with different voice parameters. Use the free trial credit or your balance to generate samples. Check the response for the audio file in your chosen format. Review the audio to ensure the tone and speed match your expectations.
What is the watermark?
Every generated audio file contains an inaudible watermark that marks it as synthetic speech. This watermark is embedded in the audio data and is not audible to listeners. It serves to identify the audio as AI-generated without affecting the listening experience or quality.
Female Voices
For no-code authors, selecting the right female voice sets the emotional baseline for your audiobook. Our API provides four distinct presets, each designed for specific narrative styles without requiring complex configuration.
Aria: Warm narrator
Aria delivers a clear, warm tone ideal for standard narration. Her voice remains consistent over long texts, making her a reliable default for most manuscripts.
- Ideal for general fiction and memoirs
Nova: Bright and upbeat
Nova offers a bright, energetic delivery. Use her voice to bring energy to dialogue or lighter narrative passages where a more vibrant tone is desired.
- Best for lively dialogue or contemporary fiction
Luna: Soft storytelling
Luna provides a softer, more intimate sound. This preset works well for atmospheric scenes or first-person narratives that require a gentle, personal touch.
- Suitable for introspective or atmospheric scenes
Sage: Calm and instructional
Sage features a calm, steady pace. She is particularly effective for non-fiction, self-help books, or instructional content where clarity and composure are prioritized.
- Optimal for non-fiction and educational content
Male Voices
The male voice options provide distinct tonal ranges suitable for various audiobook styles, from energetic marketing to deep documentary narration. Each preset is derived from licensed recordings, ensuring consistent quality across your manuscript without requiring custom voice design.
These eight presets cover a wide spectrum of narrative needs. You can select the voice that best matches the mood of your text, adjusting speed and language parameters to fit the specific requirements of your project.
Felix: Energetic promo
Felix delivers an energetic delivery style ideal for promotional chapters or fast-paced segments. His tone is lively and engaging, making him suitable for content that requires high energy levels.
- Energetic tone
Milo: Friendly conversational
Milo offers a friendly, conversational style that mimics natural speech patterns. This voice is well-suited for personal narratives or books that aim to connect directly with the listener on an informal level.
- Conversational tone
Atas: Neutral news
Atlas provides a neutral, news-style narration. His clear and balanced delivery works well for factual content, biographies, or non-fiction where clarity and objectivity are prioritized over emotional expression.
- Neutral tone
Orion: Deep documentary
Orion features a deep voice profile, often associated with documentary-style storytelling. His rich tone adds gravity and seriousness to historical accounts or dramatic narratives.
- Deep tone

Genre Matching
Matching voice presets to genre helps set listener expectations without subjective interpretation. The eight available presets allow you to select a tone that aligns with your manuscript’s structure.
Romance voices
Romance narratives benefit from warmth and clarity. Aria provides a clear, warm narration style suitable for emotional scenes. Luna offers a soft, storytelling quality that supports intimate passages.
- Aria: Warm, clear narrator for dialogue-heavy scenes
- Luna: Soft, storytelling tone for emotional depth
Thriller voices
Thrillers require tension and depth. Orion’s deep, documentary-style voice builds suspense effectively. Felix adds energy and urgency for action sequences, keeping listeners engaged.
- Orion: Deep, documentary tone for suspense
- Felix: Energetic, promo style for high-stakes moments
Non-fiction voices
Non-fiction demands clarity and neutrality. Atlas delivers a neutral, news-style delivery ideal for factual content. Sage provides a calm, instructional tone for educational material.
- Atlas: Neutral, news-style for factual accuracy
- Sage: Calm, instructional for educational content
Tone Analysis
Tone analysis begins with selecting the appropriate voice preset based on the text's inherent rhythm. Each voice has a distinct acoustic profile, ranging from warm and clear to deep and documentary-style. The choice determines the baseline emotional context without requiring additional production work.
Voice characteristics
Select a voice preset that aligns with your manuscript’s pacing. The eight available voices provide different tonal baselines, such as warm narration or energetic delivery. These presets are fixed; you cannot modify pitch or add emotion tags, so choose the preset closest to your desired tone before generating.
Speed adjustments
Adjust the speed parameter between 0.25 and 4.0 to control the delivery rate. Standard narration typically runs at 1.0. Faster speeds reduce generation time but may reduce clarity, while slower speeds enhance emphasis for complex passages. Test a 100-character sample to find the optimal listening pace.
Narrative style
Match the voice’s natural cadence to your genre. Dialogue-heavy sections benefit from conversational presets, while descriptive passages may suit calmer voices. Since there is no SSML, you must rely on punctuation and paragraph breaks in your input text to guide the natural pauses and emphasis.
Technical Specs
Audio output is available in multiple formats including mp3, wav, opus, flac, and pcm. Each file contains an inaudible watermark to identify synthetic speech. There is no defined bitrate tier or noise floor metric; quality depends on the selected format and compression settings.
Audio quality
Choose the format that best fits your distribution channel. MP3 offers small file sizes for web delivery, while WAV and FLAC provide lossless audio for archival purposes. PCM is suitable for direct integration into audio editing software.
- Formats: mp3, wav, opus, flac, pcm
- Inaudible watermark on every file
- No specific bitrate or dB metrics defined
Latency
Generation speed is roughly real-time, processing about 25 characters of text per second of audio. Short phrases may take a few seconds to synthesize. Requests are processed sequentially, so heavy load may cause queuing delays.
- ~25 characters per second of audio
- Sequential processing under load
- Short phrases: few seconds delay
Consistency
The same voice preset maintains a consistent tone across all generations. Since each request is processed independently, background noise or tone shifts are minimal. Long texts should be split into chunks of 4096 characters to maintain control.
- Consistent voice tone per preset
- Independent request processing
- Split long texts for control
Comparison Table
Selecting the right narrator requires matching voice traits to your manuscript’s structure. Each of the eight presets offers a distinct tonal baseline, allowing you to maintain consistent character voices or narrator styles across chapters.
Voice characteristics
Voices are categorized by gender and tonal quality. Female voices include Aria (warm narrator), Nova (bright, upbeat), Luna (soft, storytelling), and Sage (calm, instructional). Male options are Felix (energetic, promo), Milo (friendly, conversational), Atlas (neutral, news), and Orion (deep, documentary).
Best use cases
Use Aria or Luna for fiction requiring emotional depth. Nova suits light non-fiction or self-help. Sage is ideal for instructional content. Felix works for promotional chapters. Milo suits conversational memoirs. Atlas fits neutral non-fiction. Orion suits deep narratives or documentaries.
- Fiction: Aria, Luna
- Non-fiction: Nova, Sage, Atlas
- Promo: Felix
- Memoir: Milo
- Documentary: Orion
Limitations
Voices are static presets; you cannot adjust emotion or pitch beyond the 0.25–4.0 speed range. There is no SSML support for fine-tuning pauses or emphasis. Each voice maintains a consistent tone throughout, which may not suit stories requiring dramatic tonal shifts within a single sentence.
Testing Strategy
To ensure your audiobook sounds natural, start by generating short samples of dialogue or narration. This helps you verify that the voice choice matches your intended tone before committing to a full chapter.
Short samples
Generate 50–100 character segments to test specific emotional tones or character voices. This low-cost approach lets you iterate quickly without spending significant credits on long passages that might not fit your vision.
Long text tests
Process a full chapter (up to 4096 characters) to check for consistency and pacing. Since requests are processed sequentially, longer texts may queue, so test during low-load periods to gauge realistic latency.
- Check for abrupt stops or unnatural pauses at paragraph breaks.
- Verify that speed settings (0.25–4.0) maintain clarity over long durations.
Multi-language tests
Generate audio in each target language to verify flow and pronunciation. Since the API returns only audio files, you must listen manually to confirm natural cadence, as pronunciation is not explicitly validated by the system.
- Test non-Latin scripts (e.g., Arabic, Japanese) to ensure correct rendering.
- Set the
languageparameter explicitly for Latin-script texts to avoid default English guessing.

Voice Licensing
Our voices are derived from the LibriTTS-R dataset, which consists of public-domain audiobook readings from LibriVox. We use these licensed recordings as the foundational source material for our synthesis models. This ensures that the core vocal characteristics are rooted in established, accessible literary recordings.
LibriTTS-R source
The LibriTTS-R dataset provides high-quality speech samples that form the basis of our eight narrator voices. These recordings are carefully selected to ensure clarity and natural flow, suitable for long-form content like audiobooks. By leveraging this specific dataset, we maintain consistent audio quality across all generated files.
- Based on LibriVox public-domain readings
- High-quality speech samples
- Designed for long-form content
CC BY 4.0
The source recordings are licensed under CC BY 4.0. This license applies to the original source material rather than the specific synthetic voice models we generate. You do not need to attribute the original LibriVox readers when using our API outputs, as the license covers the input data, not the final synthetic result.
- CC BY 4.0 on source material
- License covers input data, not output
- No attribution required for final audio
Public domain
Many of the source readings come from public-domain works, ensuring broad usability for commercial audiobooks. This reduces legal friction when publishing your final audio files. You can confidently use the generated speech for your projects without worrying about complex copyright restrictions on the underlying voice data.
Customization
Customize output parameters to match your production workflow. Adjust playback speed and select the audio container that fits your distribution channel without requiring external conversion tools.
Speed control
Set the speed parameter between 0.25 and 4.0. This allows you to slow down dense non-fiction for clarity or speed up light fiction for a brisk pace. The API processes requests sequentially, so extreme speeds may affect latency during high load.
- Range: 0.25x to 4.0x
Format selection
Choose from mp3, wav, opus, flac, or pcm via the response_format field. There is no inherent quality difference between these formats; they differ only in compression and file size. Select mp3 or opus for efficient storage, or wav/flac for lossless archival.
- Supported: mp3, wav, opus, flac, pcm
Language override
Use the optional language field with an ISO code to force a specific language. If omitted, the system guesses based on script, defaulting to English for Latin text. Explicitly setting the language prevents misidentification for non-English Latin-script content.
- Optional ISO code parameter
Common Pitfalls
Choosing the wrong voice or misconfiguring settings is the most common reason an audiobook sounds unnatural. This guide helps you avoid those errors by matching voice characteristics to your genre and ensuring technical consistency across chunks.
Wrong tone choice
Matching a voice to a genre is subjective. A warm narrator may feel too soft for high-energy promo, while a deep documentary voice might drag in a fast-paced thriller. Test samples in context, not in isolation.
- Aria suits romance but may lack energy for action scenes.
- Felix is energetic but may feel too promotional for serious non-fiction.
Inconsistent pacing
Pacing issues usually stem from chunking or queueing, not speed settings. Requests are processed sequentially, so long texts may experience variable latency. Ensure your splitter joins audio seamlessly without leaving gaps or overlaps.
- Split text into parts of up to 4096 characters.
- Join audio files carefully to maintain narrative flow.
Language mismatch
Unsupported languages like Vietnamese or Thai will likely produce garbled audio if not explicitly handled. Even supported languages may default to English if the script is Latin. Always set the language field for non-English Latin text to ensure correct pronunciation.
- Explicitly set
languagefor non-English Latin-script text. - Unsupported languages will not render correctly.
Final Selection
Choosing the right narrator voice for your audiobook requires balancing your manuscript’s tone with the specific capabilities of the TTS engine. You must evaluate whether the voice’s inherent character matches the story’s emotional arc, keeping in mind that all eight voices support any of the 23 available languages. This ensures the final audio feels cohesive from start to finish, regardless of the genre.
Since the API processes requests sequentially, longer texts may experience queue delays, so consistency in pacing is vital. Select a voice that maintains clarity and emotional stability over extended listening sessions. Avoid voices that might fatigue the listener after hours of continuous playback, ensuring the audio remains engaging throughout the entire manuscript.
Decision criteria
Prioritize voices that match your manuscript’s core tone. For example, use Aria for warm narratives or Orion for deep documentaries. Ensure the selected voice remains clear and engaging over long durations, as the API processes requests sequentially and may queue under load. Verify that the voice supports the target language without requiring manual overrides, unless your text uses non-standard scripts.
- Match voice tone to genre (e.g., Luna for soft storytelling, Felix for energetic promo)
- Verify language support for all chapters
- Check speed settings (0.25–4.0) for optimal pacing
- Ensure 4096-character limit per request is manageable for your text split strategy
Audience preference
Test short samples with your target readers to gauge reception. Since voices are based on licensed recordings from LibriTTS-R, the tone is fixed; you cannot adjust emotion tags or add custom pauses. Use the free trial credit ($0.50) to generate varied samples and let your audience choose the most natural-sounding narrator. This objective feedback loop helps avoid subjective assumptions about voice appeal.
- Generate samples using the $0.50 free trial
- Ask readers to rate clarity and engagement
- Test with both short phrases and longer excerpts
- Compare multiple voices side-by-side
Brand alignment
Ensure the voice’s character aligns with your publishing brand’s identity. For instance, a professional non-fiction publisher might prefer Atlas for its neutral, news-like tone, while a creative fiction imprint might choose Nova for a bright, upbeat feel. Since the audio carries an inaudible synthetic watermark, the brand remains distinct from human narrators. This clarity helps maintain transparency with your audience regarding the audio’s origin.
- Align voice personality with brand values
- Maintain consistency across all titles
- Use the inaudible watermark to denote synthetic speech
- Choose voices that reflect your brand’s tone

Next Steps
Before committing to full production, generate short audio samples for each chapter using your preferred voice. This allows you to verify that the tone matches your manuscript’s style and that the audio quality meets your standards.
Generate samples
Use the API endpoint to convert small text excerpts into audio files. Test different formats like MP3 or WAV to ensure compatibility with your publishing platform. Keep samples under the 4,096 character limit to keep costs predictable.
Get feedback
Share your samples with a small group of beta readers or editors. Ask them to evaluate clarity, pacing, and emotional resonance. This external review helps identify if a voice feels too monotone or mismatched to the story's genre.
Start production
Once you select the best voice, split your manuscript into chunks of 4,096 characters or less. Generate each part sequentially and join the audio files. Monitor your API key usage to ensure you stay within your balance limits during this phase.
- Split text into 4,096-character segments
- Generate audio for each segment
- Join files using a simple script
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Turn your written words into immersive audio stories without writing a single line of code today
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- LibriVox public-domain audiobooks
- OpenAI API reference: create speech
- MDN: Web Speech API
- FFmpeg documentation
- W3C WAI: making audio and video accessible
- IETF RFC 7845: Ogg Opus
- Node.js documentation
- Python documentation
- LibriTTS-R paper (arXiv 2305.18802)
- TRON developer hub: TRC-20
- openai-python on GitHub
