WaveRNN
E200566
audio waveform generator
autoregressive neural vocoder
neural network architecture
text-to-speech vocoder
WaveRNN is a neural network-based audio waveform generator designed as a more efficient, real-time alternative to earlier autoregressive models for tasks like text-to-speech synthesis.
All labels observed (3)
| Label | Occurrences |
|---|---|
| WaveRNN canonical | 2 |
| dual softmax WaveRNN | 1 |
| sparse WaveRNN | 1 |
How this entity was disambiguated
This entity first appeared as the object of triple T1793248 — resolving that mention is where its identity was fixed. The disambiguator weighed these candidate entities and picked the highlighted one (or “None”, minting a new entity). This is how homonymy is resolved: the same surface form can point to different entities.
NED1
Entity disambiguation (via context triple)
gpt-5-mini-2025-08-07
Target entity: WaveRNN Context triple: [WaveNet, ledTo, WaveRNN]
-
A.
WaveNet
WaveNet is a deep generative neural network architecture for raw audio that produces highly natural-sounding speech and other audio signals.
-
B.
Versoix
Versoix is a Swiss municipality on the shores of Lake Geneva, known as a residential suburb of Geneva with lakeside promenades and a mix of urban and natural landscapes.
-
C.
GPT-2
GPT-2 is a large transformer-based language model known for generating coherent, human-like text and sparking widespread discussion about the implications of advanced AI text generation.
-
D.
Audion
Audion is an early triode vacuum tube invented by Lee de Forest that enabled the amplification of electrical signals and was crucial to the development of radio and electronics.
-
E.
GPT-3
GPT-3 is a large-scale autoregressive language model known for generating human-like text and performing a wide range of natural language tasks with minimal fine-tuning.
- F. None of above. chosen
- G. Unsure - the case is ambiguous/there is not enough information to decide.
NED2
Entity disambiguation (via description)
gpt-5-mini-2025-08-07
Target entity: WaveRNN Target entity description: WaveRNN is a neural network-based audio waveform generator designed as a more efficient, real-time alternative to earlier autoregressive models for tasks like text-to-speech synthesis.
-
A.
WaveNet
WaveNet is a deep generative neural network architecture for raw audio that produces highly natural-sounding speech and other audio signals.
-
B.
Versoix
Versoix is a Swiss municipality on the shores of Lake Geneva, known as a residential suburb of Geneva with lakeside promenades and a mix of urban and natural landscapes.
-
C.
GPT-2
GPT-2 is a large transformer-based language model known for generating coherent, human-like text and sparking widespread discussion about the implications of advanced AI text generation.
-
D.
Audion
Audion is an early triode vacuum tube invented by Lee de Forest that enabled the amplification of electrical signals and was crucial to the development of radio and electronics.
-
E.
GPT-3
GPT-3 is a large-scale autoregressive language model known for generating human-like text and performing a wide range of natural language tasks with minimal fine-tuning.
- F. None of above. chosen
Statements (49)
| Predicate | Object |
|---|---|
| instanceOf |
audio waveform generator
ⓘ
autoregressive neural vocoder ⓘ neural network architecture ⓘ text-to-speech vocoder ⓘ |
| basedOn | recurrent neural network ⓘ |
| belongsTo |
autoregressive generative models
ⓘ
neural vocoder family ⓘ |
| comparedTo | WaveNet ⓘ |
| designedFor |
audio waveform generation
ⓘ
real-time speech synthesis ⓘ text-to-speech synthesis ⓘ |
| hasAdvantage |
faster-than-WaveNet generation
ⓘ
reduced computational cost per audio sample ⓘ suitable for deployment on edge devices ⓘ |
| hasArchitectureType | RNN ⓘ |
| hasComponent |
coarse-fine 16-bit sample modeling
ⓘ
dual softmax output layer ⓘ single-layer gated recurrent unit ⓘ sparse recurrent matrix (in some variants) ⓘ |
| hasKeyProperty |
autoregressive sample-level generation
ⓘ
computational efficiency ⓘ high-quality audio synthesis ⓘ low-latency synthesis ⓘ real-time inference on CPUs ⓘ |
| hasTrainingObjective |
cross-entropy loss on audio samples
ⓘ
maximum likelihood estimation ⓘ |
| hasVariant |
WaveRNN
self-linksurface differs
ⓘ
surface form:
dual softmax WaveRNN
WaveRNN self-linksurface differs ⓘ
surface form:
sparse WaveRNN
|
| improvesOn | WaveNet ⓘ |
| inputType |
acoustic features
ⓘ
mel-spectrograms ⓘ |
| introducedAs | efficient neural audio synthesis model ⓘ |
| moreEfficientThan |
PixelCNN-based vocoders
ⓘ
WaveNet ⓘ |
| outputDomain | time-domain audio samples ⓘ |
| outputType | raw audio waveform ⓘ |
| researchArea |
deep learning for audio
ⓘ
generative modeling of waveforms ⓘ speech synthesis ⓘ |
| supports |
high sampling rates
ⓘ
real-time generation at 24kHz (on suitable hardware) ⓘ |
| targetHardware |
CPU
ⓘ
GPU ⓘ mobile devices ⓘ |
| usedFor |
neural text-to-speech systems
ⓘ
neural vocoding ⓘ speech synthesis research ⓘ |
| usedIn |
end-to-end TTS pipelines
ⓘ
neural speech synthesis toolkits ⓘ |
How these facts were elicited
The pipeline generated the facts above by prompting gpt-5.1 with this entity's name + description and the instruction below.
Instruction
You are a knowledge base construction expert. Given a subject entity and a description of it, return factual statements that you know for the subject as a JSON list of dictionaries(triples), where keys must be "subject", "predicate" and "object". The number of facts may be very high, between 25 to 50 or more, for very popular subjects. For less popular subjects, the number of facts can be very low, like 5 or 10. # Requirements - If you don't know the subject at all, return an empty list. - If the subject is not a named entity, return an empty list. - Include at least one triple where predicate is "instanceOf". - Do not get too wordy. - Separate several objects into multiple triples with one object.
Input
Subject: WaveRNN Description of subject: WaveRNN is a neural network-based audio waveform generator designed as a more efficient, real-time alternative to earlier autoregressive models for tasks like text-to-speech synthesis.
Referenced by (4)
Full triples — surface form annotated when it differs from this entity's canonical label.
this entity surface form:
sparse WaveRNN
this entity surface form:
dual softmax WaveRNN