Assume that the following sentence - "Hi, how are you doing today?" was spoken by…
- User
Assume that the following sentence - "Hi, how are you doing today?" was spoken by someone and then transcribed using a speech to text system. What transcription errors could be introduced into the final transcribed text?
Assistant
In this example, several transcription errors could occur, including:
-
Phonetic errors - This type of error results from the transcription program not being able to correctly recognize the speech patterns in the sentence. For example, if the speaker pronounced the word "doing" with a long "o" sound, the transcribed text might read "doin'."
-
Omissions - This type of error occurs when the speech-to-text program fails to transcribe an entire word or phrase. For example, if the speaker paused briefly before saying "today," the transcribed text might read "how are you doing today."
-
Insertions - This type of error occurs when the speech-to-text program transcribes an extra word or phrase that was not actually spoken. For example, if the speaker pronounced the word "doing" with a short "o" sound, the transcribed text might read "Hi, how are you doing today?"
-
Word-level errors - This type of error occurs when the speech-to-text program transcribes a word incorrectly. For example, if the speaker pronounced the word "doing" with a long "o" sound, the transcribed text might read "Hi, how are you doing today?"
-
- User
Based on the above guidelines, can you corrupt my example to show such errors in action?
Assistant
Here are a few examples of corrupted speech-to-text transcriptions of your sentence:
-
Phonetic errors: Hi, how are you doin' today?
-
Omissions: Hi, how are you doing today?
-
Insertions: Hi, how are you doing today?
-
Word-level errors: Hi, how are you doing today?
-
- User
Can you create some more examples for all of the above errors except for phonetic errors?
Assistant
Sure! Here are a few more examples of corrupted speech-to-text transcriptions for omissions, insertions, and word-level errors:
- Omissions:
- Hi, how are today?
- Hi, how you doing today?
- Insertions:
- Hi, how are you doing today, how are you doing today?
- Hi, how are you doing today, hello?
- Word-level errors:
- Hi, how are you bing today?
- Hi, how are hearing you doing today?
Note that these are just a few examples, and there are many other ways that the speech-to-text program could corrupt the transcription.
It's also worth noting that many modern speech-to-text programs use machine learning algorithms, and as a result, they often get better at transcribing speech as they are trained on more and more data. This means that the chances of these types of errors decreases as the system is trained on a larger and more diverse set of audio data.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.