Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Microsoft AI / NaturalSpeech

NaturalSpeech

Original single-speaker English text-to-waveform research system, using a variational autoencoder with duration and prior/posterior modeling.

Speech & voice
restrictedClosed weightsText-to-speech
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
13 of 26 specification fields collected
Audio sample rates22.05 kHzSee specification notes below
Reported latency0.013 real-time factor on one NVIDIA V100 GPU, batch size one sentence, averaged over the paper’s test sentences.See specification notes below
Input billingNot collectedMicrosoft Research (prerecorded samples)
Output billingNot collectedNo inferred price

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

No public service contract or verified value for this original research checkpoint.

Maximum input tokens
Not collected

No public service contract or verified value for this original research checkpoint.

Maximum output
Not collected

No public service contract or verified value for this original research checkpoint.

Parameters
28.7M parameters used during inferenceSource for Parameters (opens in a new tab)

Counts the phoneme encoder, differentiable durator, bidirectional prior/posterior module and waveform decoder. Not the full training network.

Reasoning
Not collected

No public service contract or verified value for this original research checkpoint.

Tool calling
Not collected

No public service contract or verified value for this original research checkpoint.

Structured output
Not collected

No public service contract or verified value for this original research checkpoint.

Streaming
Not collected

No public service contract or verified value for this original research checkpoint.

Batch processing
Not collected

No public service contract or verified value for this original research checkpoint.

Fine-tuning
Not collected

No public service contract or verified value for this original research checkpoint.

Context caching
Not collected

No public service contract or verified value for this original research checkpoint.

Knowledge cutoff
Not collected

No public service contract or verified value for this original research checkpoint.

Release date
Not collected

No public service contract or verified value for this original research checkpoint.

Input modalities
TextSource for Input modalities (opens in a new tab)
Output modalities
AudioSource for Output modalities (opens in a new tab)
Languages
EnglishSource for Languages (opens in a new tab)

Language of the documented experiment; not an exhaustive claim of supported production languages.

Research publication
2022-05-09Source for Research publication (opens in a new tab)

First research-family disclosure on arXiv; reviewed specifications may reflect a later paper revision. Not a commercial or separate variant release date.

Endpoint availability
Research paper and prerecorded samples. No public full-model download or interactive inference service found in reviewed original materials.Source for Endpoint availability (opens in a new tab)
Voice cloning
Not supportedSource for Voice cloning (opens in a new tab)

The documented LJSpeech checkpoint is single-speaker; later NaturalSpeech zero-shot abilities are not inherited.

Reference audio
Not collected

No reference-audio inference control documented for the original single-speaker checkpoint.

Architecture
End-to-end variational autoencoder with phoneme pretraining, differentiable duration prediction, bidirectional prior/posterior modeling and memory-augmented decoding.Source for Architecture (opens in a new tab)
Training dataset
LJSpeech: approximately 24 hours, 13,100 clips from one English speaker; 12,500 training / 100 validation / 500 test clips. Phoneme encoder pretraining uses 200 million news-crawl sentences.Source for Training dataset (opens in a new tab)
Training hardware
8 NVIDIA V100 32GB GPUs; 15,000 training epochs.Source for Training hardware (opens in a new tab)
Audio sample rates
22.05 kHzSource for Audio sample rates (opens in a new tab)
Tokenizer
Grapheme-to-phoneme conversionSource for Tokenizer (opens in a new tab)
Provider-reported latency
0.013 real-time factor on one NVIDIA V100 GPU, batch size one sentence, averaged over the paper’s test sentences.Source for Provider-reported latency (opens in a new tab)

Research measurement, not a service latency commitment.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Microsoft Research (prerecorded samples)

Provider source (opens in a new tab)
API model ID
Not collected

No public API model selector.

Access
Read the paper and listen to prerecorded examples. No user-callable inference endpoint established by the reviewed project.
Regions
Not applicable to prerecorded research samples.

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

No hosted customer inference or data-retention contract verified.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Microsoft Research (prerecorded samples)

Not verified

No public service cache contract verified.

How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 14 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Classification is an evidence-based inference for the original research checkpoint: the authors provide papers and prerecorded samples, but no full original model weights or callable service in the reviewed project materials. The Microsoft Hugging Face account lookup also returns no matching model. This does not classify independent implementations or separately released codec components as closed.

Classification evidence (opens in a new tab)

Lifecycle dates

Research family first published on arXiv9 May 2022Source (opens in a new tab)

Sources reviewed