Frequently asked questions
Is the migration process complex?
It is straightforward. The API uses the same request shape as OpenAI, so you only need to change the base URL to https://api.ek33.info/v1 and update your authentication key. Most existing code works without modification, allowing you to swap providers with minimal effort.
Do I need to change voice names?
Yes, you must update your voice identifiers. We offer eight specific presets like aria, nova, and atlas. Unlike Amazon Polly’s extensive library, you will select from these eight licensed voices, which can speak any of the 23 supported languages.
Will my existing code break?
No, if you use standard HTTP or the OpenAI SDK. Our endpoint mirrors the OpenAI speech structure. You simply point your SDK to https://api.ek33.info/v1 and use your new API key. The request fields remain identical, ensuring compatibility.
How do I handle long prompts?
Split your text into chunks of up to 4096 characters. Send each chunk as a separate request and join the resulting audio files. This ensures compliance with the character limit while maintaining the full message integrity for your IVR system.
Is the audio quality comparable?
Our voices are based on licensed recordings from LibriTTS-R and LibriVox. They provide clear, natural speech suitable for call centers. Our audio is distinct and professional, with an inaudible synthetic watermark.
Can I use the same speed settings?
Yes, our speed parameter accepts values from 0.25 to 4.0, matching common TTS standards. You can adjust the playback rate precisely to fit your IVR pacing requirements without needing to rewrite your audio processing logic or adjust timing significantly.
Do I need to update SSL certificates?
No. We use standard HTTPS for all API requests. Your existing SSL configuration will work seamlessly with our endpoint. No special certificates or network changes are required on your server to establish a secure connection for text-to-speech synthesis.
Is there a downtime during switch?
No. You can run both APIs simultaneously during testing. Once verified, you simply update your code to point to our endpoint. There is no service interruption required on your end, ensuring your call center operations continue without any unexpected outages.
How do I test the new voices?
Sign up for a free trial with $0.50 credit. Use the provided curl example or SDK to generate short audio clips. Test all eight voices and 23 languages to ensure they meet your IVR standards before migrating your production traffic fully.
What if the API is slow?
Requests are processed sequentially. Under load, a request may wait in a queue. Typical speed is roughly real time. If delays occur, consider splitting long texts or staggering requests to ensure smooth audio generation for your call center workflows.
Can I keep old audio files?
Yes. Generated audio is returned in the response and not stored by us. You can archive them locally or in your cloud storage indefinitely, regardless of whether you continue using our API or not.
Do I need a new account?
Yes, create an account at /get-api-key/. Accounts are per site, so you need a separate login for ek33.info. You can sign up with Google or email. Each account gets one active API key for authentication and billing.
Is data migration automatic?
No. You must manually update your code’s base URL and API key. Voice names also need updating. There is no automatic import of your previous voice configurations or history. You control the migration timeline and verify each change yourself.
How do I verify billing?
Check your balance and usage on the dashboard. We charge $0.01 per 1,000 characters. Failed requests are free. You can top up with USDT on TRON. View detailed usage logs to track characters consumed and ensure accurate billing for your team.
Understanding the Amazon Polly Context
Amazon Polly has served as a reliable TTS engine for many IVR systems, but its ecosystem introduces layers of complexity that indie developers and small call center teams often find burdensome. While the voice quality is solid, the surrounding infrastructure can feel disproportionate to the task of simply converting text to speech.
Why developers seek alternatives to Polly
Many teams are looking for a simpler, more transparent API that doesn’t require managing multiple AWS services. The desire for a single endpoint that handles speech synthesis without complex service integrations is a primary driver. Developers want to focus on their core product rather than configuring auxiliary cloud resources.
- Reduced infrastructure overhead for small teams
- Simpler integration workflows
- Transparent, predictable pricing models
Complexity of AWS IAM and login flows
Setting up IAM roles and permissions can be time-consuming, especially for independent developers who aren’t dedicated DevOps engineers. The login flows and credential management add friction to the development cycle, requiring careful configuration to ensure secure and reliable access to the TTS service.
- IAM role configuration
- Credential management overhead
- Security setup complexity
Cost surprises in high-volume IVR
In high-volume IVR scenarios, costs can escalate unexpectedly due to per-character billing nuances and data transfer fees. Understanding the true cost of every character, including spaces and punctuation, helps in budgeting. Transparent pricing ensures that scaling up doesn’t lead to surprise invoices at the end of the month.
- Per-character billing clarity
- Data transfer fee awareness
- Predictable monthly costs
API Endpoint Differences
The Amazon Polly endpoint structure is distinct from our standardized REST approach. While Polly relies on specific AWS service paths, we simplify this to a single POST https://api.ek33.info/v1/audio/speech endpoint. This eliminates the need for complex AWS-specific URL routing, reducing the cognitive load for indie developers managing their IVR infrastructure.
Standard POST vs AWS specific paths
By consolidating the API into one clear endpoint, you remove the overhead of navigating AWS-specific paths. Our structure mirrors the widely adopted OpenAI speech endpoint, allowing you to keep your existing HTTP client logic intact while simply updating the target URL.
JSON body structure comparison
Our JSON body requires model, input, voice, response_format, and speed. This is significantly cleaner than Polly’s parameter-heavy requests. For instance, you no longer need to manage complex SSML tags; the API handles mp3, wav, opus, flac, or pcm directly within the same simple payload.
- Use
tts-1ortts-1-hdfor the model - Specify text in the
inputfield - Choose from 8 preset voices via the
voicekey
Authentication header similarities
Authentication remains straightforward with a Bearer token. You simply pass Authorization: Bearer tts-... in your headers, similar to how you might handle API keys in other modern services. This consistency makes swapping out your TTS provider a matter of changing the base URL and headers, rather than rewriting your entire request logic.
Voice Mapping Strategy
Migrating from Amazon Polly requires mapping legacy voice IDs to our eight presets. Since Polly offers dozens of neural voices, you must select the closest match for your IVR context. Our voices are designed for clarity and natural flow in call center environments, reducing the need for complex parameter tuning.
Unlike Polly’s granular SSML controls, we do not support SSML. This simplifies your codebase while maintaining high audio quality. Use the voice mapping table below to align your existing prompts with our presets.
Mapping Polly voices to our presets
Start by identifying the emotional tone of your current Polly voices. Female voices like Joanna or Salli map best to Aria, Nova, Luna, or Sage. Male voices like Matthew or Justin align with Felix, Milo, Atlas, or Orion. Test each mapping with your standard IVR prompts to ensure the tone matches your brand’s expectations.
- Joanna/Salli -> Aria or Luna
- Matthew/Justin -> Milo or Atlas
- Ivy/Kendra -> Nova
- Russell -> Felix or Orion
Aria vs Joanna: tone comparison
Aria offers a warm, clear narrator tone. While Joanna is known for its consistent, professional neutrality, Aria adds a subtle warmth that can make automated calls feel more engaging without sounding artificial.
Adjusting speed for natural flow
Polly users often adjust SSML tags like <prosody> to control pacing. Here, you use the speed parameter (0.25–4.0). Start with a speed of 1.0 and increase to 1.2–1.4 for standard IVR menus to keep callers engaged. Avoid speeds above 2.0, as clarity drops significantly. Test with native speakers to find the optimal balance between speed and comprehension.
Handling Text Length Limits
IVR menus often exceed the 4096-character limit per request. You must split long prompts into smaller chunks, ensuring each segment stays within the boundary.
Generate audio for each part independently, then concatenate the files before playing them to the caller. This keeps latency low and handles complex scripts.
Splitting long IVR prompts
Break your script into segments under 4096 characters. Count spaces and punctuation as characters. Send each part as a separate POST request to the speech endpoint.
- Keep segments concise to avoid queue delays.
- Preserve natural pause points between chunks.
Joining audio chunks seamlessly
Combine the resulting audio files in the correct order. Most IVR systems support concatenating WAV or MP3 streams. Ensure consistent sample rates to prevent audio glitches.
Managing 4096 character boundaries
Monitor the character count strictly. Requests exceeding the limit may fail or truncate. Use your dashboard to track usage and ensure you stay within the per-request constraint.
- Count all spaces and punctuation.
- Test boundary cases with your longest menu.
Testing Voice Quality
Before migrating your IVR system, validate that the synthetic voices meet your quality standards without risking your main balance. The free trial provides $0.50 in credit, allowing you to generate approximately 50,000 characters over seven days. This window is ideal for testing edge cases, such as unusual punctuation or specific brand names, to ensure the TTS engine handles them correctly before committing to a full rollout.
Using the free trial for validation
Use your trial credit to generate test prompts for your most critical call flows. Since the API processes requests sequentially, monitor how voice quality holds up under typical load. This low-risk environment lets you verify that the audio quality meets your team's expectations before you replace your existing provider.
Comparing audio formats (MP3 vs WAV)
Choose MP3 for standard IVR menus to save bandwidth, or select WAV if your telephony system requires uncompressed PCM audio. The API supports both formats, along with Opus, FLAC, and raw PCM. Ensure your selected format matches the codec requirements of your PBX or SIP trunk to avoid playback issues in the customer's ear.
- MP3: Best for bandwidth efficiency and general IVR use.
- WAV: Ideal if your PBX requires uncompressed audio.
- Opus/FLAC: Available for specialized high-fidelity needs.
Checking latency in call flows
TTS generation is not instantaneous; long texts take time proportional to their length, roughly real-time. Short phrases may take a few seconds to process. When testing, measure the delay between sending the request and receiving the audio file. Ensure this latency does not cause awkward pauses that disrupt the natural flow of your IVR menu.
- Short phrases: A few seconds generation time.
- Long texts: Approximately real-time speed (25 chars/sec).
- Queueing: Requests may wait under high load.
Updating Code Libraries
Migrating from Amazon Polly to our API requires minimal code adjustments. Our endpoint mimics the OpenAI speech structure, allowing you to swap the base_url in your existing OpenAI SDK clients to https://api.ek33.info/v1 without rewriting your integration logic.
Switching base_url in OpenAI SDK
Replace the standard OpenAI base URL with our API endpoint. This single change directs your SDK requests to our infrastructure while maintaining the same JSON payload structure for model, input, voice, and response format parameters.
- Update base_url to https://api.ek33.info/v1
- Keep existing OpenAI SDK client configuration
- Use the same authentication header format (Bearer token)
Minimal code changes required
Since we accept the same request shape as the OpenAI speech endpoint, your IVR routing logic remains intact. You only need to ensure your new API key is correctly passed in the Authorization header.
Retesting integration endpoints
After updating the URL, verify that your IVR system correctly receives the audio file. Test with short prompts first to confirm latency, then scale to longer menus to ensure seamless chunking if needed.

Monitoring Usage and Costs
Tracking your character consumption is straightforward. The dashboard on /get-api-key/ displays your remaining balance, characters left, and usage history in real time. Since billing is based strictly on input characters (including spaces and punctuation), you can predict costs precisely without guessing.
Our flat rate of $10 per 1 million characters simplifies forecasting. This transparency helps indie developers and small call center teams avoid unexpected bills during high-volume IVR migrations.
Tracking characters in the dashboard
The account page provides a clear view of your current balance and character usage. You can also see your trial credit status if you signed up with Google. This visibility allows you to monitor consumption without needing complex third-party tools.
Since failed requests are free, only successful synthesizations count against your limit. This means you can test edge cases without worrying about inflating your usage numbers incorrectly.
- Real-time balance updates
- Character count includes spaces and punctuation
- Failed requests are free
Comparing cost per call
With a fixed price of $0.01 per 1,000 characters, calculating cost per call is simple. Multiply your average prompt length by this rate to estimate per-call expenses. This predictability is crucial for budgeting in small call center environments.
For example, a 200-character IVR prompt costs roughly $0.002. There are no hidden costs for API calls, just the raw character usage.
Setting up alerts for balance
While we do not offer email or SMS alerts directly, you can monitor your balance via the dashboard. Since the base rate is transparent, you can set your own internal thresholds using simple scripts that check the /v1/voices endpoint or your account page.
Keep an eye on your key usage. If you exceed 60 requests per minute, you might hit rate limits, but your balance remains accurate regardless of queue delays.
Dealing with Language Settings
When migrating from Amazon Polly, explicit language codes prevent accidental misidentification. Our API guesses language from script, but Latin text defaults to English. For non-English Latin text, always set the language parameter to ensure accuracy.
This precision matters for multilingual IVR menus where a single misheard word changes meaning. By defining the language explicitly, you avoid fallback errors and ensure the voice matches the caller's expected dialect.
Explicit language code usage
Set the language field in your JSON body to an ISO code. This overrides automatic detection, ensuring consistent output regardless of input script nuances.
Without this setting, Latin-script text might default to English, causing pronunciation errors in languages like Dutch or Portuguese.
- Supported: Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish
Handling multilingual IVR menus
For IVRs switching between languages, pass the correct language code with each request. This ensures the engine selects the appropriate phonetic rules.
Since any voice can speak any supported language, you can maintain a single voice preset (e.g., Aria) across multiple languages for brand consistency.
Fallback voice selection
If a specific language isn't supported, you must handle the error gracefully. Our API does not support Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, or Bengali.
Use a static audio file as a fallback for unsupported languages, or switch to a voice known for clarity, like Atlas, to minimize confusion.
Error Recovery Procedures
Legacy IVR systems often crash when TTS services hang. We ensure your call flow remains robust by treating synthesis errors as recoverable events rather than fatal failures. This keeps your customers hearing something, even if it isn’t the dynamic voice.
Handling API failures gracefully
When the API returns an error, the response is free, so you lose no credit. Your code should catch these HTTP errors immediately and trigger a fallback path. This prevents the caller from listening to silence while your system decides what to do next.
- Check for 4xx or 5xx status codes in your integration layer
- Log the error reason (e.g., invalid character limit, network timeout)
- Ensure the call does not hang indefinitely waiting for a response
Retry logic for network issues
Network glitches happen. Implement a simple retry mechanism with exponential backoff for transient errors. Since requests are processed sequentially, a brief pause between retries helps avoid hitting the 60 requests-per-minute limit again too quickly.
- Retry once or twice for 503 Service Unavailable errors
- Increase wait time between retries (e.g., 1s, 2s, 4s)
- Stop retrying after a set timeout to avoid blocking the call thread
Fallback to static audio files
If dynamic synthesis fails repeatedly, play a pre-recorded MP3 of the same message. This is crucial for critical prompts like account numbers or PIN entry. Keep these static files lightweight and ensure they match your brand’s tone.
- Store common prompts as static MP3 or WAV files
- Map specific error codes to predefined fallback audio files
- Test fallback audio clarity and volume levels regularly
Final Verification Steps
Before going live, verify the entire call flow works with the new API. Place test calls through your IVR to ensure audio plays correctly and transitions smoothly between prompts.
This step confirms that the synthesized speech integrates seamlessly with your existing telephony infrastructure without unexpected delays or errors.
Running end-to-end call tests
Initiate a full call cycle using the actual IVR menu structure. Verify that voice prompts trigger at the correct decision points and that the audio connects without significant latency.
Ensure that the transition from user input to speech response feels natural for callers.
- Test menu navigation paths
- Verify hold music integration
- Check timeout behaviors
Validating audio clarity
Listen to the generated audio in your target environment. Ensure the voice quality is clear and the speed settings (0.25–4.0) match your IVR pacing requirements.
Confirm that the audio format (mp3, wav, etc.) is compatible with your telephony provider's requirements.
- Check volume levels
- Verify format compatibility
- Assess voice tone fit
Confirming billing accuracy
Review the dashboard to ensure character counts match your expectations. Verify that the cost per call aligns with your budget calculations based on the $0.01 per 1,000 characters rate.
Ensure that failed requests were correctly excluded from your balance to avoid overcharging.
- Check character count logs
- Verify cost per call
- Confirm balance updates
Support and Resources
We keep things simple because your call center doesn’t need bureaucracy. Our documentation covers every migration step, from endpoint configuration to voice mapping. You can find ready-to-use code examples and clear explanations of our pricing model without digging through complex AWS IAM policies.
If you get stuck during the switch, our team is here to help. Support is by email only.
Documentation link access
All technical details are available at /docs/. This includes the full API reference, voice descriptions, and pricing breakdowns. The documentation is updated regularly to reflect current features and limits.
You can also test the API directly using the curl examples provided on /get-api-key/. This allows you to verify your setup before making changes to your production code.
- Full API reference
- Voice descriptions and pricing
- Curl examples
Email support for migration help
For specific migration questions, contact us via email. We focus on helping indie developers and small teams navigate the transition from Amazon Polly to our simpler API.
Our support team can assist with troubleshooting integration issues, clarifying voice options, or explaining character counting for your billing needs.
Community forums and guides
While we don’t have a dedicated community forum, we encourage developers to share their migration experiences and code snippets. This helps others in the indie developer community understand how to adapt their IVR flows.
Check our /about/ page for more background on our team and our commitment to transparent, honest TTS solutions.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Her gentle cadence helps complex information land clearly, reducing the need for callers to ask for repetition during onboarding or support calls.
23 languages
Any voice speaks any of these languages.
