Frequently asked questions
What is synthetic speech?
Synthetic speech is audio generated by our TTS API when you send text via a POST request. Each file carries an inaudible watermark marking it as synthetic. The input text is used only to produce the audio and is not used for training models, ensuring your content remains distinct.
How is it different from human voice?
Unlike human recordings, our voices are presets based on licensed recordings from LibriTTS-R and LibriVox. Every generated file includes an inaudible watermark identifying it as synthetic speech. While natural, it lacks human emotion tags or custom voice cloning. It offers consistent, clear narration ideal for repetitive language learning exercises.
Is it used for training models?
No. Input text is used only to produce the immediate audio file and is not used for training models. We do not store generated audio, only request metadata for billing. This ensures your lesson content remains private and distinct from any broader model training data.
Can it sound natural?
Voices like Aria and Nova provide clear, consistent narration suitable for lessons. While natural, they lack emotion tags and custom pronunciation lexicons. The speed is roughly real-time, and requests are processed sequentially, so complex texts may queue under load, but the output remains clear and consistent.
How many languages does it support?
We support 23 languages, including English, Spanish, German, French, Japanese, and Arabic. Any voice can speak any supported language. If the language parameter is omitted, the system guesses from the script, defaulting to English for Latin text. Vietnamese and Thai are not supported.
Is it expensive for video content?
Pricing is $10 per 1 million characters, or $0.01 per 1,000 characters. Spaces and punctuation count. Failed requests are free. There are no monthly plans or minimums. This flat rate allows you to generate thousands of practice tracks predictably without hiring voice actors for each lesson.
Can I use it commercially?
Yes. The API provides a straightforward infrastructure for generating audio assets. Voices are based on licensed recordings. You pay per character used. The API provides a straightforward infrastructure for generating audio assets. Balance does not expire, allowing you to build a library of lesson tracks over time.
How long does it take to generate?
Generation is roughly real-time, processing about 25 characters of text per second. Short phrases take a few seconds. Requests are processed sequentially, so under load, a request may wait in a queue. Longer texts are split into parts and joined automatically to ensure consistent delivery.
Does it have accents?
Voices are presets with specific tones, such as Aria’s warm narration or Felix’s energetic promo style. They do not have configurable accents. You can set the language parameter to guide pronunciation. If omitted, Latin text defaults to English. Accents are inherent to the voice preset, not adjustable.
Can I change the speed?
Yes, you can adjust the speed between 0.25 and 4.0 using the speed parameter. This allows you to create slower practice tracks for beginners or faster ones for advanced learners. The default speed produces audio at roughly real-time, but you can customize this for your specific lesson needs.
Is there a character limit?
Each request accepts up to 4,096 characters of text. For longer lessons, split the text into parts and join the resulting audio files. This limit ensures consistent processing times. Failed requests within this limit are free, so you can experiment with different voice settings without cost.
Do I need technical skills?
You need basic API knowledge to send a POST request with JSON body. The endpoint mimics the OpenAI speech format, so existing OpenAI SDKs work by changing the base URL. You can also use the ready curl example on the dashboard. No official SDK of our own is required; you can use the OpenAI SDK or plain HTTP.
Can I hear a sample?
You can hear samples by generating audio via the API. Use the /v1/voices endpoint to list available presets like Aria, Nova, or Felix. Each voice has a distinct tone, from soft storytelling to energetic promo. Generate a short test phrase to hear the specific quality for your lessons.
How do I pay for it?
Top up using USDT on the TRON network (TRC-20), minimum $5. Credits are automatic within minutes. No cards or PayPal. Start with a $0.50 trial if you sign in with Google. Balance does not expire. Failed requests are free, and you pay only for successful generations.
Definition of Synthetic Voice Generation
Synthetic voice generation converts written text into audible speech using algorithmic models. For language learning creators, this API provides a reliable infrastructure to produce thousands of practice tracks without hiring voice actors. The service acts as a cloud-based endpoint where input text yields standardized audio files.
The system processes requests in JSON format, allowing seamless integration into existing educational workflows. By decoupling voice production from recording studios, creators can maintain consistent audio quality across entire course libraries while focusing on pedagogical content rather than scheduling sessions.
Text-to-Speech Technology Explained
Our API delivers synthesized audio via simple HTTP requests. Users send text and receive an audio file, eliminating the need for physical recording equipment. This method ensures that every lesson sounds professionally produced, regardless of scale.
The technology supports 23 languages, enabling educators to generate materials in Arabic, Chinese, or Swahili with the same ease as English. Each request processes up to 4096 characters, providing substantial content per call.
- Cloud-based endpoint: POST request returns audio
- Supports 23 languages including Spanish and Japanese
- No recording studio required
Algorithmic Voice Synthesis
The engine uses licensed recordings from LibriTTS-R and LibriVox public-domain audiobook readings. These sources provide a base of natural speech patterns that the algorithm replicates to create distinct voice presets like Aria, Nova, and Felix.
Each generated file carries an inaudible watermark marking it as synthetic. This ensures transparency for learners while maintaining the clarity needed for language acquisition exercises.
- Based on LibriTTS-R (CC BY 4.0)
- Public domain audio base
- Inaudible synthetic watermark included
Digital Audio Production
Audio is generated in formats like MP3, WAV, Opus, FLAC, or PCM. This flexibility allows direct integration into video editors or learning management systems without conversion steps. Speed can be adjusted between 0.25x and 4.0x for different learning levels.
Requests are processed sequentially, taking roughly real-time duration for long texts. Short phrases may take a few seconds, ensuring high-fidelity output suitable for pronunciation drills.
- Formats: MP3, WAV, Opus, FLAC, PCM
- Adjustable speed: 0.25–4.0x
- Sequential processing ensures quality
Benefits for Language Educators
Synthetic voice generation allows language educators to maintain a consistent auditory experience across their entire curriculum. By using the same voice preset for all exercises, learners build stronger neural associations with specific phonetic patterns, reducing cognitive load when switching between lessons.
This consistency is difficult to achieve with rotating human voice actors. Our API ensures that every word in a beginner’s course sounds as if it were spoken by the same instructor, reinforcing learning through predictable prosody and clear articulation.
Consistency Across Lessons
When students hear the same voice across multiple modules, they develop better recognition of subtle accent differences and intonation patterns. This uniformity supports long-term retention and reduces the mental effort required to adjust to new speakers in each new lesson.
Using presets like Aria or Felix ensures that every practice track maintains identical pacing and tone, which is critical for auditory learners who rely on consistent audio cues to distinguish between similar-sounding words in target languages.
- Single voice preset across entire course modules
- Uniform prosody for better phonetic recognition
- Reduced cognitive load for learners
Instant Content Updates
Educators can update lesson text and regenerate audio in minutes without waiting for studio availability. If a pronunciation example needs correction or a cultural reference changes, you simply update the JSON input and receive a new audio file immediately.
This eliminates the delays of booking voice talent, scheduling recording sessions, and editing raw audio files. You retain full control over the exact wording and timing of each practice track, ensuring accuracy without logistical bottlenecks.
- Regenerate audio in minutes, not days
- No studio booking or scheduling delays
- Immediate correction of errors or updates
Scalability for Large Libraries
Generate thousands of practice tracks simultaneously using the API’s 60 requests per minute limit. Each request processes up to 4096 characters, allowing you to create extensive libraries of vocabulary, dialogues, and reading exercises without hiring additional talent.
The $0.01 per 1,000 characters rate makes scaling cost-effective. You can produce hours of unique audio content for a fraction of the cost of human recording, enabling the creation of diverse, language-specific content for any of the 23 supported languages.
- Supports high-volume batch generation
- Cost-effective at $0.01 per 1,000 characters
- No limit on the number of voices or languages
Multi-Language Support Capabilities
SonicLingo’s API supports 23 languages, enabling creators to generate practice tracks for diverse learner populations without hiring multiple voice actors. The system covers major European, Asian, and Middle Eastern languages, including Arabic, Chinese, Japanese, and Swahili.
We provide consistent voice profiles across all supported languages. If a learner encounters the same voice in English and French modules, the auditory experience remains familiar, reducing cognitive load and aiding recognition.
23 Supported Languages
The API supports Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish. Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, and Bengali are not supported.
- Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish
Cross-Lingual Voice Consistency
Any of the 8 voice presets can speak any of the 23 supported languages. This allows you to maintain a single character identity across different language modules, helping students build stronger auditory associations.
The language parameter is optional but recommended for non-English Latin-script text. If omitted, the system guesses the language, defaulting to English for Latin characters, which may cause mispronunciation for other languages.
Accent Accuracy in TTS
Voices are based on licensed recordings from the LibriTTS-R dataset (CC BY 4.0) and LibriVox public-domain audiobooks. This foundation ensures clear, standard pronunciation suitable for educational contexts, though specific regional accents are not configurable.
Since voices are synthetic, they lack the subtle emotional nuances of human speech but offer consistent clarity ideal for repetition-based language drills.
Comparison with Human Voice Actors
Human voice actors provide emotional nuance but require significant time and budget for studio booking, editing, and revisions. For language learning creators, hiring talent to record thousands of practice tracks can become a logistical bottleneck, especially when lesson content needs frequent updates or expansion into new languages.
Cost Efficiency Analysis
Our API charges $0.01 per 1,000 characters, making it significantly cheaper per minute of audio than traditional voiceover rates. Failed requests are free, and there are no monthly retainers or minimum spend requirements. You only pay for the text you actually generate.
- No per-hour studio fees
- No waiting for talent availability
- Pay-as-you-go with USDT top-ups
Speed of Production
Generate hours of audio in minutes rather than days. While human recording requires scheduling, soundproofing, and post-production, our API processes requests sequentially. Longer texts are split and joined automatically, allowing you to scale your library instantly.
- Instant generation for short phrases
- Automated splitting for long texts
- No post-production editing needed
Variety vs. Human Nuance
Human actors offer unique emotional depth, but our eight preset voices (e.g., Aria, Felix, Atlas) provide consistent, clear tones ideal for educational clarity. The synthetic nature is marked by an inaudible watermark, ensuring transparency while maintaining high audio quality for learners.
Technical Implementation in Video
SonicLingo’s API returns audio files ready for direct insertion into video editing timelines, eliminating the need for intermediate rendering steps. This streamlined workflow allows language learning creators to generate thousands of practice tracks efficiently, ensuring consistent audio quality across extensive lesson libraries without hiring voice actors.
Audio-Video Synchronization
Generated audio aligns precisely with video frames, supporting accurate timing for reading exercises and pronunciation guides. Since requests are processed sequentially, longer texts may experience slight queue delays, but the output remains stable for synchronous playback in educational videos.
- Direct audio file delivery for easy timeline insertion
- Consistent timing for reading aloud exercises
- Supports standard video editing workflows
File Format Compatibility
The API supports MP3, WAV, OPUS, FLAC, and PCM formats, ensuring compatibility with most learning management systems and video editors. Choose the format that best balances file size and quality for your specific distribution channel, whether it’s web-based lessons or downloadable resources.
- MP3, WAV, OPUS, FLAC, and PCM support
- Optimize for web or offline distribution
- No proprietary format restrictions
Integration with Learning Platforms
Use the standard OpenAI-compatible SDK or plain HTTP requests to integrate SonicLingo into your existing platform. Change the base URL to https://api.hajee.info/v1 and use your API key to generate audio on demand, automating content creation for video lessons at scale.
- OpenAI SDK compatible
- Standard HTTP requests
- Easily integrates with LMS platforms
Data Privacy in Synthetic Audio
Synthetic speech generation prioritizes data integrity for educational content. Input text is processed solely to produce the audio output and is not used for training models. This ensures your lesson scripts remain proprietary and are not ingested into public datasets.
No Training on User Text
Your input text is used exclusively to generate the audio file and is not used for training. This means your lesson content, scripts, or custom exercises are not ingested into the model's training data.
Temporary Audio Processing
Generated audio is returned directly in the API response and is not stored on our servers after delivery. Only the request metadata—such as timestamp, character count, and status—is retained for billing purposes.
Metadata Retention Policies
Request metadata is kept strictly for billing and usage tracking. No long-term storage of the actual audio content or source text occurs, ensuring your educational materials remain yours.
Quality and Realism Expectations
SonicLingo uses licensed recordings from LibriTTS-R and LibriVox to generate synthetic voices. The resulting audio features natural prosody, ensuring that stress and rhythm mimic human speech patterns for clearer language acquisition.
Each generated file carries an inaudible watermark, marking it as synthetic speech. This helps distinguish our TTS output from human-recorded lessons, maintaining transparency for learners and educators alike.
Natural Prosody in TTS
The voices exhibit natural intonation patterns, avoiding the robotic monotone common in older TTS systems. This helps learners grasp the musicality of the target language, such as the tonal variations in Mandarin or the stress patterns in English.
- Consistent stress and rhythm across all 8 voices
- Natural phrasing without awkward pauses
- Suitable for conversational and narrative content
Clarity for Language Learners
Clarity is prioritized for educational use. Voices like Aria and Luna offer clear, distinct articulation, ideal for pronunciation drills. Others, like Felix, provide energetic delivery suitable for interactive exercises. The API supports 23 languages, allowing consistent voice quality across different linguistic contexts.
- Clear articulation for pronunciation practice
- Support for 23 languages
- Optional speed adjustment (0.25–4.0x) for difficult passages
Watermarking Synthetic Audio
Every audio file generated by our API contains an inaudible watermark. This digital signature identifies the audio as synthetic, which is useful for copyright tracking and distinguishing our content from human-recorded materials in educational libraries.

Common Use Cases in Education
To support language acquisition, educators can generate consistent audio materials for students. Our API enables the creation of extensive libraries of practice tracks without the cost or scheduling constraints of hiring voice actors.
Reading Aloud Exercises
Generate MP3 or WAV files from any text up to 4096 characters. This allows learners to hear accurate pronunciation of articles, stories, or vocabulary lists instantly.
- Support for 23 languages, including less common ones like Swahili and Malay.
- Instant generation ensures rapid content updates for lesson plans.
Pronunciation Guides
Use specific voices like Aria or Felix to model clear, standard speech patterns. The synthetic voice provides a consistent baseline for students to compare against their own attempts.
- Adjustable speed (0.25–4.0) helps learners parse fast or slow speech.
- Clear enunciation aids in distinguishing similar phonemes across different languages.
Conversational Practice
Create dialogue tracks by alternating between different voice presets. This simulates natural conversation flow, allowing students to practice listening and response timing in a controlled environment.
- Combine voices like Milo (conversational) and Atlas (news) for varied tones.
- No need for complex SSML; simple text input suffices for basic dialogues.
Future of TTS in Language Learning
As technology evolves, expect even more nuanced speech synthesis. Our platform remains adaptable, ensuring that educators can integrate new voice models as they become available.
Limitations of Current TTS Tech
The synthetic voices deliver clear, consistent pronunciation ideal for language drills, but they lack the emotional range of human actors. You cannot tag specific feelings like sarcasm or excitement; the tone remains fixed based on the chosen preset, whether it is the warm narrator Aria or the energetic Felix.
While the clarity aids learners, the system has hard limits on input length and processing speed that affect workflow. Understanding these constraints helps educators plan their content generation strategy effectively.
No Emotion Tags Available
Our API does not support emotion tags or SSML markup to modulate tone dynamically. Each voice preset has a fixed emotional character, such as the calm instructional style of Sage or the upbeat nature of Nova. If a lesson requires a sudden shift in mood, you must select a different voice preset for that segment rather than adjusting a single voice's emotional output.
Character Limit Constraints
Each API request accepts a maximum of 4,096 characters. For longer lessons, you must split the text into multiple requests and join the resulting audio files. Spaces and punctuation count toward this limit, so concise phrasing helps maximize content per request. This constraint means you cannot paste an entire textbook chapter into a single call.
Processing Queue Delays
Requests are processed sequentially, meaning your audio may wait in a queue if many users are generating content simultaneously. While typical speed is roughly real-time, short phrases can take a few seconds to generate. Under load, delays increase, so plan for batch processing during off-peak hours to ensure timely delivery of your lesson assets.
How Voices Are Created
The synthetic voices powering SonicLingo originate from licensed recordings, ensuring legal clarity for educational content. We do not claim ownership of the underlying human performances, but we license their use for synthesis.
This approach allows us to offer eight distinct voice presets while maintaining a transparent chain of custody for the audio data used in your lessons.
Licensed Recording Sources
Our voice models are built on recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. This ensures that the audio you generate is safe for public distribution in video lessons.
LibriTTS-R Dataset
The core dataset is LibriTTS-R, which provides high-quality, clean speech data. This dataset forms the backbone of our voice synthesis algorithms, delivering clear articulation ideal for language learning.
Public Domain Audio Base
Many of our source recordings come from LibriVox, a collection of public-domain audiobook readings. This base allows us to offer diverse voices like Aria and Felix without excessive licensing costs, keeping prices low for creators.
Choosing the Right Voice Model
Selecting the right voice model starts with choosing between tts-1 and tts-1-hd. Both use the same engine and pricing ($0.01 per 1,000 characters), so the choice depends entirely on your audio quality preference. The tts-1-hd model offers slightly higher fidelity, which can be beneficial for nuanced language lessons where subtle phonetic distinctions matter.
The eight available voices—Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion—cover a range of tones from warm narration to energetic promos. Since every voice supports all 23 languages, you can maintain consistent auditory branding across your entire course library. Choose a voice that matches your lesson’s energy level to keep learners engaged.
tts-1 vs tts-1-hd Differences
Both tts-1 and tts-1-hd utilize the same underlying engine and incur identical charges. The primary difference lies in audio fidelity. For standard language exercises, tts-1 is sufficient. If your lessons rely on precise phonetic clarity, tts-1-hd may offer a marginal improvement in sound quality.
The pricing structure remains uniform: $10 per 1 million characters. There is no penalty for choosing higher quality, making tts-1-hd a viable option for all content types without budget complications.
- Same engine, same price ($0.01/1k chars)
- Identical character limits (4096 per request)
- Choose
tts-1-hdfor slightly better fidelity
Selecting Based on Tone
Your choice of voice should align with the pedagogical goal. Use Sage for calm, instructional grammar breakdowns or Luna for soft storytelling. For energetic vocabulary drills, Felix or Nova provide a bright, upbeat tone that keeps energy high.
Since any voice can speak any of the 23 supported languages, you do not need to switch voices for different languages. This allows you to maintain a consistent teacher persona across diverse lessons, which aids learner recognition and comfort.
- Use
Sagefor instructional content - Use
Felixfor energetic drills - Any voice works for any supported language
Testing Voices Before Bulk Use
Before generating thousands of tracks, use the free trial credit ($0.50) to test voice consistency. Generate short phrases in the target language to verify clarity and tone. Check if the voice handles specific phonemes clearly for your audience.
Use the GET /v1/voices endpoint to review available voices and their descriptions. Pair this with small test batches to ensure the synthetic output meets your educational standards before committing to larger production runs.
- Test with the $0.50 trial credit
- Verify phoneme clarity in target languages
- Use
GET /v1/voicesto preview descriptions
Future of TTS in Language Learning
The API infrastructure currently processes requests sequentially, meaning generation speed depends on queue load rather than true real-time streaming. While long texts approach real-time playback speed, short phrases may experience slight delays under heavy load.
Real-Time Generation Potential
Current architecture handles requests in a queue, processing them one after another. While longer texts generate at roughly real-time speed, shorter phrases may take a few seconds depending on server load. True WebSocket streaming of partial audio is not yet available.
Improved Naturalness
Voices are built from licensed LibriTTS-R recordings, providing consistent prosody and clarity ideal for educational repetition. While they lack emotional tags or custom pronunciation lexicons, the natural flow supports clear listening practice.
Broader Language Support
The API currently supports 23 languages, including major European and Asian tongues. It does not yet support Vietnamese, Indonesian, Thai, or Ukrainian. Future expansion depends on dataset availability rather than algorithmic changes, keeping the focus on high-quality, consistent voice presets.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. This is critical for beginners who are still developing their auditory processing skills for new sounds and rhythms.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- IETF RFC 7845: Ogg Opus
- W3C WAI: making audio and video accessible
- OpenAI API reference: create speech
- openai-python on GitHub
- Python documentation
- LibriVox public-domain audiobooks
- LibriTTS-R corpus (OpenSLR 141)
- MDN: Web Speech API
- TRON developer hub: TRC-20
- openai-node on GitHub
- LibriTTS-R paper (arXiv 2305.18802)
