Audio sample ratesHistorical whitepaper: 22.05 kHz input to 44.1 kHz output. A future minimum 16 kHz to 48 kHz version is described as upcoming. August 2022 article instead states 16.05 kHz to 44.1 kHz; current supported rates are unresolved.See specification notes below
LanguagesNot collectedAwaiting source review
Input billingNot collectedRespeecher audio enhancement service (historical)
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 model-specific specification verified; unknown is not unsupported.
Maximum input tokens
Not collected
No model-specific specification verified; unknown is not unsupported.
Maximum output
Not collected
No model-specific specification verified; unknown is not unsupported.
Parameters
Not collected
No model-specific specification verified; unknown is not unsupported.
Release date
Not collected
No model-specific specification verified; unknown is not unsupported.
Knowledge cutoff
Not collected
No model-specific specification verified; unknown is not unsupported.
Tool calling
Not collected
No model-specific specification verified; unknown is not unsupported.
Structured output
Not collected
No model-specific specification verified; unknown is not unsupported.
Streaming
Not collected
No model-specific specification verified; unknown is not unsupported.
Batch processing
Not collected
No model-specific specification verified; unknown is not unsupported.
Fine-tuning
Not collected
No model-specific specification verified; unknown is not unsupported.
Reasoning
Not collected
No model-specific specification verified; unknown is not unsupported.
Context caching
Not collected
No model-specific specification verified; unknown is not unsupported.
Base model
Not collected
No model-specific specification verified; unknown is not unsupported.
Creation privacy
Website policy retains personal data as necessary, with legal exceptions and international transfers. Its separate AI Voice Assistant appendix promises no model training or voice cloning for assistant conversations and specifies 12-month transcript/technical-data retention. Those assistant-only terms are not a Space TTS or Marketplace inference-retention guarantee.Source for Creation privacy (opens in a new tab)
GAN-based neural audio enhancer trained on high-resolution recordings paired with artificially downsampled versions. Generates missing spectral content; this is not proof of recovering the exact original waveform.Source for Architecture (opens in a new tab)
Audio sample rates
Historical whitepaper: 22.05 kHz input to 44.1 kHz output. A future minimum 16 kHz to 48 kHz version is described as upcoming. August 2022 article instead states 16.05 kHz to 44.1 kHz; current supported rates are unresolved.Source for Audio sample rates (opens in a new tab)
Documentation discrepancy
Whitepaper and article give different input-rate coverage. Super48 is mentioned in the whitepaper but no separate released checkpoint is identified. Noise reduction, dereverberation, decompression and bit-depth enhancement are roadmap items there, not proven shipped capabilities.Source for Documentation discrepancy (opens in a new tab)
Published price
Not collected
Historical contact-based service; no current public numeric price or documented API verified.
No immutable checkpoint selector verified; voice IDs identify catalogue profiles.
Access
Historical whitepaper offers contact-based service delivery. No current self-service application, public API or weight download verified.
Regions
Inference residency not verified; company location is not processing-region evidence.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Service-specific terms and exact inference retention require confirmation. Website policy retains personal data as necessary, with legal exceptions and international transfers. Its separate AI Voice Assistant appendix promises no model training or voice cloning for assistant conversations and specifies 12-month transcript/technical-data retention. Those assistant-only terms are not a Space TTS or Marketplace inference-retention guarantee.
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.
Respeecher audio enhancement service (historical)
Not verified
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
How it works
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
Minimum reusable input
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
Lifetime & refresh
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
How to enable it
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
Usage fields
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
Billing rules
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
Limits & data boundaries
Not collected
No inference-cache contract verified. WebSocket connection reuse, continuing prosody context, deterministic seeds and saved recordings do not establish discounted input caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 16 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
Provider identifies its own GAN-based enhancement network, originally internal and later offered as a service. Reviewed article and whitepaper publish no downloadable weights; this is a historical model family, not a verified current checkpoint or delivered roadmap version.
