Custom Avatar Model Training Showing as Processing After 16 Hours

Question

Custom Avatar Model Training Showing as Processing After 16 Hours

Trinanjan Majumder 0 Microsoft Employee

I created a Azure AI Service Resource in West US 2(Test Avatar) and then went to Speech Studio, uploaded all the required training Data and then started the model training. But its showing 1hr left estimated for last 8 Hours.

Jerald Felix 10,975 Reputation points

2026-02-25T02:45:38.7333333+00:00
Hello Trinanjan Majumder,

Thanks for raising this question in Azure Q&A forum.

A custom avatar model training staying stuck in "Processing" status is a known issue with Azure AI Speech Studio, and there are a few common causes and fixes to try.

First, it's worth knowing that custom avatar training is a resource-intensive process it normally takes 20–40 compute hours on average depending on the amount of training data you've provided. If your training was submitted recently, give it at least a full day before concluding it's genuinely stuck.

However, if it has been over 48 hours with no progress or status change, it's likely stuck. Here are the things to check and try:

Check your training data quality. The most common reason avatar training hangs in processing rather than outright failing is that the system encounters frames where the avatar talent's face is not clearly detected across every frame of the video. The training engine uses strict facial landmark detection and even subtle issues like motion blur, hand movements near the face, inconsistent lighting, or slight head turns can cause the backend to stall. Review all your gesture and training videos carefully before re-submitting.

Check your region. There have been documented cases where specific regions (like West US) had backend issues processing avatar training requests. If you're in a region that has had recent Azure AI service stability problems (such as Sweden Central which had major outages in late January 2026), try creating a new Speech resource in a different region like West Europe or East US, and submit the training there.

Verify your Speech resource tier. Custom avatar training only works with a Standard (S0) pricing tier for Azure AI Foundry or Speech resources it is not supported on Free (F0) tiers. If you're on a Free tier, this is likely why training never progresses.

Delete the stuck training and re-submit. If the training cannot be canceled or deleted from the Speech Studio UI, you can force-delete it using the Speech REST API:

bash DELETE https://<region>.api.cognitive.microsoft.com/speechtotext/v3.2/models/<model-id>

Get the model ID from Speech Studio under the training entry, then re-submit the training job fresh.

Open an Azure Support ticket. If none of the above resolves it, this is a backend service issue that requires Microsoft to investigate the stuck job. Open a support ticket at https://portal.azure.com/#blade/Microsoft_Azure_Support/HelpAndSupportBlade with your Speech resource name, region, training job ID, and timestamp of when it got stuck. Microsoft Support can unblock or cancel the job from the backend.

If it helps kindly accept the answer.

Best Regards,

Jerald Felix
SRILAKSHMI C 15,030 Reputation points Microsoft External Staff Moderator

2026-02-26T04:41:51.1933333+00:00

Hi Trinanjan Majumder,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!

1 answer

Your answer

SRILAKSHMI C 15,030 Reputation points Microsoft External Staff Moderator

2026-02-26T04:41:51.1933333+00:00

Hi Trinanjan Majumder,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!

Answer 1

Hello Trinanjan Majumder,

Welcome to Microsoft Q&A and Thank you for reaching out.

I understand that your Custom Avatar training has been stuck in “Processing – 1 hour left” for 8+ hours, that’s understandably frustrating. Below is a consolidated troubleshooting guide combining expected behavior, common causes, and recommended actions.

1. Understand Expected Training Time

Custom Avatar training (under Azure AI Speech) is compute-intensive.

Typical timelines:

Small/standard datasets → 2–6 hours

Larger datasets → can take 20–40 compute hours

ETA may fluctuate during processing

However, if The ETA remains fixed (e.g., “1 hour left”) for many hours

There is no visible stage progression

That may indicate backend queuing or job orchestration delay.

Please refer this How to create a custom video avatar

2. Validate Service Configuration

Please confirm The resource is a Speech resource (not OpenAI)

SKU is Standard (S0) Custom Avatar training does NOT support Free (F0)
Region (West US 2) supports Avatar training

3. Check Training Data Quality

Training can stall if the uploaded videos have issues such as:

Motion blur

Poor or inconsistent lighting

Face partially obstructed

Inconsistent framing

Audio noise or overlapping voices

Even if initial validation passed, borderline-quality datasets can slow processing significantly.

4. Region Capacity or Backend Queue

Occasionally, a region (e.g., West US 2) may experience:

Training queue congestion

Backend capacity throttling

Orchestration delays

Recommended steps:

Check Azure Service Health for West US 2

If feasible (for testing), create a Speech resource in:

East US
West Europe And attempt training there

If it completes normally in another region, this suggests regional capacity constraints.

5. Check Azure Portal Logs & Quota

Go to Azure Portal → Speech Resource → Activity Log

Look for Failed operations, Quota errors, Throttling messages

Also verify:

You have not hit concurrent training limits
Subscription quota is sufficient

6. Cancel and Retry

If cancellation is available in Speech Studio:

Cancel the training job

Wait ~15–30 minutes

Restart training

If it immediately returns to the same stuck state, that indicates backend pipeline delay rather than dataset issues.

7. Force Delete

If Training exceeds 24–48 hours, No stage change, No error surfaced

You can delete the model via REST API if the UI does not allow it:

DELETE https://<region>.api.cognitive.microsoft.com/speechtotext/v3.2/models/<model-id>

You will need:

Region name

Model ID from Speech Studio

Proper authentication

After deletion, recreate and retrain.

Please refer this Train model guide

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Share via

Custom Avatar Model Training Showing as Processing After 16 Hours

1 answer

Your answer