Pricing and usage of Pronunciation Assessment feature from Azure Speech-to-Text

Question

I am trying to use Pronunciation Assessment feature for my app. I will be assessing a given audio file instead of streaming live,

To my understanding, the default pronunciation assessment pricing is same as "Standard Transcription : Real-time Transcription = $1/hr" even if I request asynchronously with an existing audio file

I also found out that for audio files shorter than 30 seconds, I can use Speech to text REST API for short audio to save costs. I'm not sure if the price is still at $0.66/hr for this

https://learn.microsoft.com/en-us/answers/questions/5574771/can-i-use-the-azure-speech-to-text-fast-transcript
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/rest-speech-to-text-short

Lastaly, I found out that the Python speech SDK has a "recognize_once_async" operation that transcribe utterances of up to 30 second which also has the Pronunciation Assessment feature.

With all the informations above I have 2 questions.

1. Are my pricing informations correct?

• Pronunciation Assessment : $1/hr
• Speech to text REST API for short audio : $0.66/hr

1. Is the Python speech SDK's "recognize_once_async" operation the same as "Speech to text REST API for short audio"? It seems like the same feature, but is the pricing also the same?

Answer 1

Hello YJ Kim,

Your understanding of Azure AI Speech pricing for Pronunciation Assessment and short audio transcription is mostly correct, but with some updates based on the latest details—note that rates can vary by region, tier, and date (as of November 2025, US East pricing applies; always check the Azure pricing calculator for your setup). I'll break it down by question, confirming costs and features for asynchronous audio file processing (under 30 seconds, as mentioned).

1. Pricing Confirmation

Yes, your info aligns closely with current pay-as-you-go (PAYG) rates, but Pronunciation Assessment isn't a separate tier—it's billed under Speech to Text Standard. Here's the accurate breakdown:

• Pronunciation Assessment: $1.32 per hour (or ~$0.000367 per second) for Standard real-time transcription, which applies to asynchronous file processing too. This includes evaluation of audio files via SDK or REST API (e.g., for your app assessing pre-recorded speech). No extra charge for the assessment feature itself—it's the same as base Speech to Text since it builds on transcription. For short files (<30s), costs are prorated in 1-second increments, so an 8-second assessment might cost ~$0.0029.
• Speech to Text REST API for Short Audio: $0.66 per hour (or ~$0.000183 per second) for the fast transcription endpoint (Speech to text REST API V3.2+ or 2024-05-15-preview). This is for synchronous processing of short audio files (up to 60 seconds, or 30 for pronunciation), returning results faster than real-time. It's cheaper than standard real-time/batch because it's optimized for quick, limited-use cases like your app—billed in 1-second increments.

Notes on Both:

• Free tier: Up to 5 audio hours/month shared across Standard/Custom Speech to Text (no batch support in free.
• Commitment tiers: Lower rates for high volume (e.g., $1/hr drops to $0.66/hr overage after 2,000 hours/month).
• Batch transcription (for longer files) is also $1.32/hr but asynchronous—use for files >60s to avoid short-audio limits
• No separate "fast transcription" discount beyond the short audio API; your linked doc confirms it's for cost savings on quick jobs.

2. Python Speech SDK's "recognize_once_async" vs. REST API for Short Audio

They're essentially the same feature under the hood but accessed differently, with matching pricing:

• Similarity: Yes, recognize_once_async in the Python Speech SDK performs single-shot, asynchronous recognition for audio up to ~30 seconds (ideal for pronunciation assessment on short utterances/files). It uses the same backend as the Speech to Text REST API for short audio, supporting the same endpoints and features (e.g., pronunciation scoring via pronunciation_assessment_config). Both are synchronous in output (quick results) and handle WAV/MP3/etc. formats up to 60s total, with pronunciation limited to 30s.
• Pricing: Identical—falls under the short audio REST API rate of $0.66/hr if using the optimized short endpoint (specify in SDK config, e.g., via SpeechConfig with short audio URL). If it defaults to standard real-time, it bills at $1.32/hr—set the endpoint explicitly to /speechtotext/v3.2/shortaudio or later for the lower rate. The SDK doesn't add overhead; it's just a wrapper around REST calls

Recommendation: For your app (async file assessment <30s with pronunciation), use the Python SDK's recognize_once_async with a short audio endpoint config for $0.66/hr savings. Example snippet:

import azure.cognitiveservices.speech as speechsdk

speech_config = speechsdk.SpeechConfig(subscription="your_key", region="your_region")

speech_config.endpoint_id = "your_short_audio_endpoint"  # e.g., for fast/short mode

audio_config = speechsdk.audio.AudioConfig(filename="your_audio.wav")

pron_config = speechsdk.PronunciationAssessmentConfig(grading_system=speechsdk.PronunciationAssessmentGradingSystem.HundredMark, granularity=speechsdk.PronunciationAssessmentGranularity.Word)

recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config, pronunciation_assessment_config=pron_config)

result = recognizer.recognize_once_async().get()  # Blocks until done

print(result.pronunciation_assessment_result)  # Scores, etc.

This keeps costs low without REST API calls directly. Test in the free tier first, and monitor via Azure portal metrics. If volumes grow, consider commitment tiers for further savings.

If your audio exceeds 30s or needs batching, switch to full REST API for $1.32/hr. Let me know specifics like region/audio format for more tweaks!

Best Regards,

Jerald Felix