Redacting portions of text transcriptions generated from inverse text normalization

1. A system, comprising:

at least one processor; and

a memory, storing program instructions that when executed by the at least one processor, cause the at least one processor to implement a machine transcription system, configured to:

obtain text data generated from one or more automated speech recognition techniques for redaction;

cause application of a number of inverse text normalization models to the text data to generate normalized text data based, at least in part, on a predicted normalization by a first inverse text normalization model for the text data;

cause application of a machine learning model, trained to recognize text for redaction, to identify one or more portions of the normalized text data for redaction;

redact the identified one or more portions of the normalized text data; and

provide the redacted text data to a destination.

2. The system of claim 1 , wherein the text data is obtained to perform a batch transcription job and wherein the destination is a data storage system specified by the batch transcription job.

3. The system of claim 1 , wherein the text data is a stream of text data for a streaming transcription job, and wherein the machine transcription system is further configured to apply a scout network to determine a number of words in the text data upon which to perform the application of the machine learning model to identify the one or more portions of the normalized text data for redaction.

4. The system of claim 1 , wherein the machine transcription system is a machine transcription service offered by a provider network that performs a received transcription job request that includes the text data on behalf of a client of the provider network.

5. A method, comprising:

obtaining, by a machine transcription system, text data for redaction;

applying, by the machine transcription system, a number of inverse text normalization models to the text data to generate normalized text data based, at least in part, on a predicted normalization by a first inverse text normalization model for the text data;

applying, by the machine transcription system, a machine learning model, trained to recognize text for redaction, to the normalized text data to identify one or more portions of the normalized text data for redaction;

redacting, by the machine transcription system, the identified one or more portions of the normalized text data; and

providing, by the machine transcription system, the redacted text data to a destination.

6. The method of claim 5 , wherein the text data is obtained to perform a batch transaction job and wherein the destination is a data storage system specified by the batch transaction job.

7. The method of claim 5 , wherein the text data is a stream of text data and wherein the method further comprises applying a scout network to determine a number of words in the text data upon which to perform the applying of the machine learning model to identify the one or more portions of the normalized text data for redaction.

8. The method of claim 5 , wherein the application of the machine learning model is performed to evaluate the normalized text data for one or more redaction classifications specified in a request to the machine transcription system.

9. The method of claim 5 , wherein the first inverse text normalization model is a neural network-based inverse text normalization model.

10. The method of claim 9 , wherein applying the number of the one or more inverse text normalization models to the text data to generate normalized text data comprises:

determining that a normalization confidence score for the predicted normalization of the text data satisfies confidence criteria; and

responsive to the determination, applying a finite state transducer to the predicted normalization to produce the normalized text data.

11. The method of claim 5 , wherein applying the number of the one or more inverse text normalization models to the text data to generate normalized text data comprises applying a finite state transducer to the predicted normalization to produce the normalized text data.

12. The method of claim 5 , wherein the text data is obtained as the output of an automatic speech recognition system.

13. The method of claim 5 , wherein the applying of the machine learning model to the normalized text data to initially identifies a further one or more portions of the normalized text data for redaction and where the method further comprises not applying redaction to the further one or more portions of the normalized text data according to a determination that a confidence score for a classification for the further one or more portions is below a confidence threshold.

14. One or more non-transitory, computer-readable storage media, storing program instructions that when executed on or across one or more computing devices cause the one or more computing devices to implement:

receiving text data for redaction;

causing application of a number of one or more inverse text normalization models to generate normalized text data based, at least in part, on a predicted normalization by a first inverse text normalization model for the text data;

causing application of a machine learning model, trained to recognize text for redaction, to the normalized text data to identify one or more portions of the normalized text data for redaction;

redacting the identified one or more portions of the normalized text data; and

providing the redacted text data to a destination.

15. The one or more non-transitory, computer-readable storage media of claim 14 , wherein the text data is received as part of performing a batch transaction job and wherein the destination is a data storage system specified by the batch transaction job.

16. The one or more non-transitory, computer-readable storage media of claim 14 , wherein the text data is a stream of text data and wherein the one or more non-transitory, computer-readable storage media store further program instructions that when executed on or across the one or more computing devices, cause the one or more computing devices to further implement applying a scout network to determine a number of words in the text data upon which to perform the applying of the machine learning model to identify the one or more portions of the normalized text data for redaction.

17. The one or more non-transitory, computer-readable storage media of claim 14 , wherein the application of the machine learning model is performed to evaluate the normalized text data for one or more redaction classification specified in a received request.

18. The one or more non-transitory, computer-readable storage media of claim 14 , wherein the first inverse text normalization model is a neural network-based inverse text normalization model.

19. The one or more non-transitory, computer-readable storage media of claim 18 , wherein, in applying the number of the the one or more inverse text normalization models to the text data to generate normalized text data, the program instructions cause the one or more computing devices to implement:

determining that a normalization confidence score for the predicted normalization of the text data satisfies confidence criteria; and

responsive to the determination, applying a finite state transducer to the predicted normalization to produce the normalized text data.

20. The one or more non-transitory, computer-readable storage media of claim 14 , wherein the one or more computing devices are implemented as part of a machine transcription service offered by a provider network.