# US 2025/0156398 A1 · System and method for correction of a query using a replacement phrase

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0156398 A1
- Document ID: US20250156398A1
- Status: Published application
- Kind: A1
- Application number: 19/020,796
- Filed: 2025-01-14
- Published: 2025-05-15
- Earliest claimed: 2019-09-05
- Assignees: SoundHound AI IP, LLC
- Applicants: SoundHound AI IP, LLC
- Inventors: Pranav Singh; Olivia Bettaglio
- Web page: https://patentgrep.com/patent/US20250156398A1

## CPC

- G06F 16/2365: {Ensuring data consistency and integrity}
- G06F 16/24522: {Translation of natural language queries to structured queries}
- G06F 40/216: using statistical methods
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/30: Semantic analysis
- G06F 40/35: Discourse or dialogue representation
- G06N 5/04: Inference or reasoning models
- G06N 7/00: Computing arrangements based on specific mathematical models
- G10L 15/01: Assessment or evaluation of speech recognition systems
- G10L 15/183: using context dependencies, e.g. language models
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

Systems and methods are provided for natural language processing using neural network models and natural language virtual assistants. The system and method include receiving a natural language phrase including a word sequence, computing corresponding error probabilities that the words are errors, and for a word with a corresponding error probability above a threshold, then computing a replacement phrase with a low error probability to provide a response from the virtual assistant depending on the replacement phrase.

## Claims

### Claim 1

A virtual assistant comprising a trained neural network model, wherein training the model uses queries to detect an error and learn to correct the error, for natural language processing to identify errors for a query presented in a natural language form, wherein the virtual assistant calculates a vector distance between a first sentiment vector and a second sentiment vector and computes a transcription error probability and a natural language understanding misinterpretation probability using the trained neural network model, wherein the transcription error probability and the natural language understanding misinterpretation probability represent an error profile for the query.

### Claim 2 (depends on 1)

The virtual assistant of claim 1, wherein probabilities are related to the vector distance and inversely related to an edit distance, and the transcription error probability exceeds the natural language understanding misinterpretation probability for large vector distances.

### Claim 3

A computer-implemented method comprising:
receiving a natural language query and transcribing it into a first word sequence;
using a statistical model, wherein training the model includes using natural language expressions identified as errors, on words within the first word sequence to compute corresponding error probabilities that the words are errors;
deriving a second word sequence having a replacement phrase for a word with a corresponding error probability above a threshold, the replacement phrase having a lower error probability.

### Claim 4 (depends on 3)

The computer-implemented method of claim 3, wherein the replacement phrase is derived from a phonetic closeness score between a candidate replacement phrase and a hypothesized erroneous phrase.

### Claim 5 (depends on 3)

The computer-implemented method of claim 3, further comprising the step of transmitting a virtual assistant query response depending on the replacement phrase.

### Claim 6 (depends on 3)

The computer-implemented method of claim 3, further comprising receiving acoustic model scores for words within the first word sequence, wherein the corresponding error probability is inversely related to the acoustic model score.

### Claim 7 (depends on 3)

The computer-implemented method of claim 3, further comprising computing a sentiment vector from the replacement phrase, wherein the virtual assistant query response depends on the sentiment vector.

### Claim 8 (depends on 3)

The computer-implemented method of claim 3, further comprising:
computing a sentiment vector from the replacement phrase;
computing the distance between the computed sentiment vector and a sentiment vector from a previous natural language query; and
determining the virtual assistant query response based on the distance being below a threshold.

### Claim 9 (depends on 3)

The computer-implemented method of claim 3, further comprising computing an error score by aggregating a plurality of error indicators.

### Claim 10 (depends on 9)

The computer-implemented method of claim 9, further comprising normalizing weighting of error indicators using the error score.

### Claim 11

A computer-implemented method comprising:
transcribing a natural language query and into a first word sequence;
using a statistical model on words within the first word sequence to compute corresponding error probabilities that the words are errors, wherein training the model includes using natural language expressions identified as errors;
deriving a second word sequence having a replacement phrase for one or more words in the natural language query where the one or more words have a corresponding error probability above a threshold.

### Claim 12 (depends on 11)

The computer-implemented method of claim 11, wherein the replacement phrase is derived from a phonetic closeness score between a candidate replacement phrase and a hypothesized erroneous phrase.

### Claim 13 (depends on 11)

The computer-implemented method of claim 11, further comprising the step of transmitting a virtual assistant query response depending on the replacement phrase.

### Claim 14 (depends on 11)

The computer-implemented method of claim 11, further comprising receiving acoustic model scores for words within the first word sequence, wherein the corresponding error probability is inversely related to the acoustic model score.

### Claim 15 (depends on 11)

The computer-implemented method of claim 11, further comprising computing a sentiment vector from the replacement phrase, wherein the virtual assistant query response depends on the sentiment vector.

### Claim 16 (depends on 11)

The computer-implemented method of claim 11, further comprising:
computing a sentiment vector from the replacement phrase;
computing the distance between the computed sentiment vector and a sentiment vector from a previous natural language query; and
determining the virtual assistant query response based on the distance being below a threshold.

### Claim 17 (depends on 11)

The computer-implemented method of claim 11, further comprising computing an error score by aggregating a plurality of error indicators.

### Claim 18 (depends on 17)

The computer-implemented method of claim 17, further comprising normalizing weighting of error indicators using the error score.
