Frequently asked questions
What makes a voice good for alerts?
Clarity and distinctiveness are key. Use Aria for clear narration or Felix for energetic promos. Ensure the voice cuts through background noise. Test with short phrases like "door unlocked." Select a voice that matches your device’s personality without being overwhelming for brief alert sounds.
Is Luna better than Nova for notifications?
Luna offers soft, storytelling tones, while Nova is bright and upbeat. For notifications, Nova’s energetic delivery often grabs attention better. Luna suits gentle reminders. Choose based on whether you want a soothing or alerting tone. Both support Spanish and other languages effectively.
Can male voices be used for smart home devices?
Yes. Atlas provides neutral, news-style clarity ideal for status updates. Milo offers a friendly, conversational tone for interactions. Orion suits deep, documentary-style prompts. All eight voices support all 23 languages, including Spanish, allowing flexible personality matching for any IoT device interface.
How does speed affect voice clarity?
Speed ranges from 0.25 to 4.0. Faster speeds may reduce clarity for complex alerts. Stick to 1.0–1.5 for clear instructions. Test your specific text at different speeds. Audio generation takes a few seconds for short phrases, so balance speed with user comprehension needs.
Should I use Spanish or English voices?
Use Spanish voices for Spanish text. The API supports 23 languages, including Spanish. If you send Spanish text with an English voice, it may sound unnatural. Set the language parameter explicitly to ensure correct pronunciation. This applies to all 8 voice presets available.
Does voice gender matter for IoT?
It depends on your brand personality. Female voices like Aria or Nova offer warmth. Male voices like Milo or Atlas provide neutrality or authority. There is no technical difference in performance. Choose based on the emotional tone you want your smart home device to convey.
How do I test voices before deployment?
Use the API endpoint to generate short audio clips. Send text via POST request to /v1/audio/speech. Review the returned audio file. Check clarity, speed, and language accuracy. Iterate with different voices and speeds until the output matches your device’s desired personality and clarity standards.
What is the best voice for instructions?
Sage is designed for calm, instructional tones, making it ideal for step-by-step guidance. Atlas offers neutral clarity for standard, no-nonsense prompts. Milo provides a friendly conversational style for more engaging interactions. All eight voices support clear articulation across 23 languages, allowing you to choose the perfect fit for your device’s personality based on your specific interface style.
Can I mix voices in one device?
Yes. Generate each audio segment separately using different voice parameters. Concatenate the resulting audio files on your device. The API returns standard audio formats like MP3 or WAV. This allows dynamic voice changes for different alerts or sections within a single device interaction.
Do voices sound robotic?
Voices are based on licensed recordings from LibriTTS-R and LibriVox. They sound natural but carry an inaudible watermark marking them as synthetic. They are not human recordings but are high-quality neural outputs.
How does language detection work?
If the language parameter is omitted, the system guesses from the script. Latin script defaults to English. For non-English Latin text, explicitly set the language parameter. Supported languages include Spanish, French, German, and 20 others. Explicit setting ensures accurate pronunciation and tone.
Is Felix too energetic for a fridge?
Felix is energetic and promo-focused. For a fridge, Milo’s friendly tone or Atlas’s neutral style might be better. Felix works well for active alerts. Test Felix with short phrases like "door open." If it feels too loud or intense, switch to a calmer voice preset.
What is the difference between tts-1 and tts-1-hd?
Both models use the same engine and price. The HD version uses the same engine and price. The API treats them identically for requests and billing. Choose tts-1-hd if you prioritize audio fidelity. Otherwise, tts-1 is fully sufficient for standard smart home alert sounds.
Can I change voices over the air?
Yes. Update the voice parameter in your API requests. Regenerate audio clips for the new voice. Deploy the updated audio files to your device via your existing OTA method. The API supports rapid regeneration, allowing quick A/B testing of different voice personalities remotely.
Voice Characteristics
Selecting a voice for smart home alerts requires balancing distinct clarity with approachable warmth. For security notifications, a crisp delivery ensures critical information is not missed, while daily reminders benefit from a tone that feels less robotic and more like a helpful household companion.
The distinction between a calm, instructional tone and an energetic, conversational one directly impacts user experience. A voice that is too loud or harsh for a simple timer alert can become irritating over time, whereas one that is too soft might be ignored during background noise.
Clarity vs Warmth
For Spanish-language smart home interfaces, clarity is paramount for security alerts. Voices like Atlas provide a neutral, news-like delivery that is clear, though they do not physically cut through ambient noise. Conversely, for daily routines like suggesting a coffee break, a warmer voice like Sage or Luna creates a more pleasant, less intrusive atmosphere that integrates smoothly into the home environment.
- Use neutral tones for urgent security notifications
- Choose warmer tones for routine daily reminders
- Ensure consistent volume levels across voice types
Energy Levels
Energy levels should match the context of the alert. High-energy voices like Felix or Nova are excellent for promotional announcements or wake-up calls, where a boost in mood is desired. However, for late-night notifications or calm instructions, lower-energy voices like Orion or Milo prevent jarring the user. Adjusting the speed parameter (0.25–4.0) allows you to fine-tune this energy without changing the voice preset.
- Match energy to alert urgency
- Use speed settings to modulate intensity
- Avoid high energy for nighttime alerts
Use Case Analysis
Smart home alerts require immediate clarity and low latency. Security notifications must cut through ambient noise, while daily reminders need a friendly, non-intrusive tone. Selecting the right voice ensures users act quickly without feeling nagged.
Security Alerts
For critical events like motion detection, use Atlas for a neutral, authoritative tone that conveys urgency without panic. Its clear enunciation ensures commands are understood even in noisy environments. Keep alert text under 4096 characters to avoid splitting, which adds processing time.
- Use Atlas for neutral, news-style clarity.
- Keep alerts under 4096 characters.
- Set speed to 1.0 for natural urgency.
Daily Reminders
Routine reminders benefit from warmer voices like Luna or Milo. Luna’s soft storytelling style makes calendar notifications feel like a gentle nudge rather than a robot command. Milo’s conversational tone suits household tasks, making the interaction feel personal.
- Use Luna for soft, storytelling reminders.
- Use Milo for conversational household tasks.
- Set speed to 0.9 for a relaxed pace.
Technical Requirements
To ensure a responsive smart home interface, developers must account for generation latency. Short alerts typically require a few seconds to synthesize, while longer texts approach real-time speeds of approximately 25 characters per second. Since requests are processed sequentially, high traffic may introduce queue delays, so your UI should reflect this processing state to users.
Response Time
Plan your interface accordingly, as the API does not offer WebSocket streaming or real-time partial audio. For immediate security alerts, keep input text concise to minimize the time between trigger and audio playback, though generation takes a few seconds. Under heavy load, a request may wait in a queue, so avoid simultaneous synthesis bursts for critical notifications.
- Short phrases take a few seconds to generate.
- Long texts process at roughly real-time speed (25 chars/sec).
- Requests are queued sequentially under load.
Audio Format
Choose the response_format that best balances quality and bandwidth for your IoT device. MP3 is standard for broad compatibility, while WAV offers uncompressed clarity. For efficient storage and transmission, OPUS and FLAC provide high quality with smaller file sizes. PCM is available for raw audio processing needs.
- Supported formats: mp3, wav, opus, flac, pcm.
- MP3 ensures broad player compatibility.
- OPUS and FLAC offer efficient compression.
Language Handling
The API automatically detects the language based on the input script, defaulting to English for Latin text. For Spanish smart home alerts, you must explicitly set the language parameter to "es" to ensure correct pronunciation, as the system cannot reliably guess Spanish from short, technical phrases.
Spanish Script Detection
Without an explicit language tag, the model may interpret Spanish words as English or neutral text, leading to awkward enunciation. Always specify language: "es" for consistent results across your IoT device’s voice commands and notifications.
- Set
languageto "es" for all Spanish text. - Avoid relying on script detection for short alerts.
Mixed Language Text
If your smart home interface mixes Spanish and English (e.g., "El sistema está en modo *eco*"), the API attempts to handle mixed scripts. However, complex code-switching may result in inconsistent pronunciation. Test edge cases with your specific voice preset before deployment.
Developer Considerations
Choosing the right voice for your smart home interface requires balancing technical constraints with user experience. For indie developers, the API’s straightforward integration allows you to test multiple voices quickly without complex SDKs. Each voice has a distinct character, from the calm instruction of Sage to the energetic promo style of Felix, helping you define your device’s personality through audio alone.
The key is matching the voice tone to the alert type. Security warnings need clarity, while daily reminders benefit from warmth. Since the API supports 23 languages including Spanish, you can ensure consistent branding across regions by explicitly setting the language tag, preventing misinterpretation of Latin-script text.
User Interface Fit
Your voice output is part of the UI. Short alerts should use voices with clear articulation like Aria or Atlas to ensure comprehension in noisy environments. Long instructions benefit from Sage’s calm, instructional tone, which reduces listener fatigue during longer commands.
The API’s simple JSON structure lets you swap voices dynamically based on context. For example, use Felix for urgent notifications and Milo for conversational feedback. This flexibility allows you to refine the user experience without waiting for software updates.
- Use Aria or Atlas for clear, concise alerts.
- Choose Sage for detailed, step-by-step instructions.
- Swap voices dynamically based on notification type.
Brand Personality
Your device’s personality should be consistent. If your brand is friendly and approachable, Milo’s conversational tone works well. For a more neutral, professional feel, Atlas provides a reliable baseline. Avoid mixing voices too frequently, as it can confuse users about whose voice they are hearing.
Since each audio file carries an inaudible watermark, you can identify synthetic speech without affecting the user experience. This helps you analyze which voices resonate best with your audience, allowing you to refine your brand’s audio identity over time.
- Maintain consistent voice tone across all notifications.
- Use Milo for friendly, conversational interactions.
- Use Atlas for neutral, professional updates.
Comparison Factors
When optimizing your smart home interface, balancing voice naturalness with processing efficiency is key. The API allows speed adjustments from 0.25x to 4.0x, letting you tighten alert cadence without losing clarity.
Character billing is straightforward: $0.01 per 1,000 characters, including spaces and punctuation. This predictable cost structure helps indie developers budget accurately for high-volume IoT notification systems.
Speed Settings
Adjusting the speed parameter directly impacts user experience. Faster settings (e.g., 2.0x) work well for urgent security alerts where brevity matters, while slower rates (0.75x) may enhance comprehension for complex instructions.
Note that requests are processed sequentially, so under load, longer or slower generations may queue. Test your specific voice and speed combination to ensure it meets your device's real-time constraints.
- Range: 0.25 to 4.0x multiplier
- Default speed is 1.0x
- Faster speeds reduce audio file size
Character Count
Each API call supports up to 4,096 characters. For longer announcements, split your text into chunks and join the audio files. This ensures consistent quality and prevents truncation.
Remember that every space and punctuation mark counts toward your 1 million character / $10 tier. Efficient phrasing can significantly extend your usable credit.
- Max 4,096 chars per request
- Spaces count as characters
- Failed requests are free
Implementation Tips
Testing voices helps you select the right tone for your smart home interface. Use the API to generate short alert samples and evaluate clarity in different environments.
Gather user feedback to refine your voice selection. Adjust settings based on real-world usage to ensure your IoT projects communicate effectively with users.
Testing Voices
Generate samples for common alerts like doorbell rings or security notifications. Compare voices like Luna for a softer approach or Felix for energetic prompts.
Use the speed parameter to test how voice pacing affects comprehension. Ensure the audio format matches your device’s playback capabilities.
- Test voice samples in various noise environments
- Adjust speed settings between 0.25 and 4.0
- Compare Luna and Felix for different alert types
User Feedback
Collect feedback from beta testers to understand which voices resonate best. This helps in fine-tuning the personality of your smart home device.
Implement a feedback loop to continuously improve the user experience. Adjust voice choices based on user preferences and usage patterns.
- Engage beta testers for real-world insights
- Refine voice choices based on user preferences
- Maintain a continuous improvement cycle

Common Mistakes
Developers often overlook how voice tone affects user trust in smart home environments. A mismatched voice can make routine alerts feel jarring or instructions feel dismissive, degrading the user experience even if the audio is technically clear.
Language handling is another frequent pitail. Assuming the system guesses correctly can lead to unnatural pronunciation, especially with mixed-language texts common in multilingual households.
Ignoring Tone
Using a high-energy voice like Felix for a gentle goodnight reminder creates cognitive dissonation. Conversely, the deep, documentary-style voice of Orion may feel too heavy for a quick kitchen timer alert. Match the emotional weight of the voice to the urgency of the notification.
Test your alert scripts with different voices. A calm voice like Sage works well for instructional updates, while Nova’s upbeat tone suits positive confirmations like "door locked."
- Avoid using high-energy voices for sleep-time notifications
- Match voice warmth to the emotional context of the alert
- Test contrast between urgent security alerts and routine reminders
Wrong Language Tag
If your text is in Spanish but the language field is omitted, the API defaults to English. This results in poor pronunciation and unnatural intonation, making the smart home interface sound less native.
Always explicitly set the language parameter to es for Spanish text. This ensures the model applies the correct phonetic rules, even if the script uses Latin characters that might otherwise default to English.
- Explicitly set
language: "es"for Spanish text - Do not rely on script detection for non-English Latin text
- Verify pronunciation accuracy after setting the language tag
Performance Metrics
Understanding generation speed and audio quality helps you set realistic expectations for your IoT devices. Requests are processed sequentially, so longer texts may queue under load. For typical use, generation runs at roughly real-time speed, producing about 25 characters of text per second of audio. Short phrases may take a few seconds to complete.
The generated audio carries an inaudible watermark marking it as synthetic speech. While not a certification, this ensures transparency for users. Audio quality is consistent across all formats, allowing you to choose the best balance between file size and fidelity for your specific hardware constraints.
Generation Speed
Requests are processed one after another, meaning under heavy load a request may wait in a queue. For long texts, generation speed is roughly real-time, averaging about 25 characters of text per second of audio. Short phrases typically take a few seconds to synthesize, ensuring a responsive experience for immediate alerts.
- Sequential processing ensures consistent quality but may introduce latency under high load.
- Typical speed: ~25 characters of text per second of generated audio.
- Short phrases take a few seconds; long texts generate at real-time speed.
Audio Quality
Every generated file includes an inaudible watermark that clearly marks the content as synthetic speech. This provides a layer of transparency for your users without affecting the audio fidelity. The quality remains consistent regardless of the format you choose, allowing you to optimize for your device's storage and bandwidth limitations.
Voice Selection Guide
Choosing the right voice for Spanish smart home alerts depends on whether you prioritize clarity for security warnings or warmth for daily reminders. The API supports all eight voices in Spanish, allowing you to match the tone to the specific context of the notification.
Aria vs Nova
Aria offers a warm, clear narration style suitable for daily reminders like weather updates or calendar alerts. Her tone is approachable and helps maintain a friendly atmosphere in the home environment without sounding overly urgent.
- Best for: Non-urgent daily updates
- Tone: Warm and clear
Felix vs Milo
Felix brings an energetic, promo-style voice suitable for time-sensitive security alerts that need to grab attention. Milo provides a friendly, conversational tone perfect for instructional prompts, such as guiding users through a setup process.
- Best for: Security alerts (Felix) or instructions (Milo)
- Tone: Energetic or conversational
Final Recommendations
For Spanish-language smart home alerts, prioritize clarity and immediate attention. Use Atlas for neutral, news-style notifications or Orion for deeper, authoritative warnings that cut through background noise. These voices ensure critical information is understood instantly without emotional distraction.
For daily reminders and instructional guidance, select voices that encourage engagement. Milo offers a friendly, conversational tone perfect for routine tasks like setting alarms or checking weather, while Sage provides a calm, instructional clarity for step-by-step device controls or setup guides.
Top Picks for Alerts
Select Atlas for standard, neutral alerts where clarity is paramount. Choose Orion if you need a deeper, more commanding presence for security warnings or urgent system notifications that require immediate user attention.
- Atlas: Neutral, news-style clarity for standard notifications.
- Orion: Deep, documentary tone for urgent security warnings.
Top Picks for Instructions
Use Milo for friendly, conversational reminders like daily schedules or weather updates. His tone feels approachable, making routine interactions feel personal. Opt for Sage for instructional content, as his calm delivery ensures users clearly understand step-by-step device controls.
Both voices support Spanish natively and handle mixed-language text well if you tag the language code. Test these voices with your specific hardware to ensure the audio format matches your device’s playback capabilities.
- Milo: Friendly and conversational for daily reminders.
- Sage: Calm and instructional for step-by-step guides.
Support
We support indie developers building smart home interfaces via email only. There is no phone number or live chat, so please include your API key and project details in your message for faster resolution.
To get started, sign up at /get-api-key/ using your Google account or an email address. This page displays your unique API key, current balance, and usage statistics. You can issue a new key at any time, which immediately invalidates the old one.
Email Contact
Our support team is available by email for all technical and billing questions. We do not offer phone support or live chat. Please provide your API key and a description of your smart home alert scenario to help us assist you efficiently.
- Email support only
- No live chat
- Include API key in queries
API Key Generation
Visit /get-api-key/ to create your account and generate your first API key. Signing in with Google grants a $0.50 trial credit valid for seven days. Accounts created with email and password start without trial credit. You can replace your key at any time.
Your key is used for authentication via the Authorization: Bearer header. Once a new key is issued, the previous one stops working immediately, ensuring secure access to your TTS API endpoints.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Even with noise-canceling features, the voice itself should remain distinct.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- OpenAI API reference: create speech
- LibriTTS-R paper (arXiv 2305.18802)
- openai-node on GitHub
- LibriTTS-R corpus (OpenSLR 141)
- Python documentation
- FFmpeg documentation
- MDN: Web Speech API
- Node.js documentation
- LibriVox public-domain audiobooks
- curl documentation
- W3C WAI: making audio and video accessible
