Understand why you Why you can't see the full transcript in Capturi
When a conversation is transcribed by Capturi's speech recognition engine, it does not result, as in many other speech recognition systems, in just one specific output, but rather in thousands of different combinations of words and phrases.
Fundamentally, Capturi recognizes words using phonetics (the sounds that make up a word) combined with knowledge of the context in which a word typically appears.
As some words are phonetically close to each other, for example, "bucket," "socket," "crocket," the speech recognizer may potentially 'consider' all three alternatives, but with different "probability/confidence."
Each of these alternatives is then combined with alternatives of other words, forming a wide range of possible sentence constructions.
The illustration provides a simplified example of how the recognizer's considerations can result in various sentence structures.
'Best path'
"Best path" is the sentence that the recognizer considers to be the most probable and is the sentence displayed in the classic transcription (which you can extract through our API, among other methods).
However, 'best path' may not necessarily be the correct sentence, and if trackers only relied on 'best path,' you might potentially miss actual correct hits that lie outside 'best path.'
Capturi's search algorithm for trackers, therefore, explores a broader range within 'best path,' increasing the likelihood that you, as a user, experience Capturi capturing the trackers you desire.
The search algorithm takes into account various factors, including the number of phonemes (sounds) in the word you have inserted as your tracker.
If we only searched in 'best path,' the so-called "recall" (the number of times a word is actually said compared to the number of times Capturi recognizes the word) would be lower. You can read more about 'recall' here. Ultimately, as a user, you might find that Capturi potentially didn't find what you wanted, leading to lower quality.
You can read more about how you can adjust the search algorithm yourself on each tracker to control how broadly it searches.
Why isn't the 'best path' transcription shown in Capturi?
The value in Capturi lies in the ease with which you, as a user/company, can quickly analyze across many conversations and automatically gain insight into the topics that are prevalent or how your employees handle specific inquiries, etc.
In this light, transcription provides no value as it only relies on one conversation. Additionally, confusion may arise between the 'best path' transcription and tracker hits in the conversation since the tracker search algorithm, as mentioned, looks broader than the 'best path' transcription.
In other words, the 'best path' transcription reflects only a small part of the recognizer output in Capturi, and therefore, it is not displayed in the conversation.
As mentioned, it is possible to retrieve the 'best path' transcription through our API if you wish to save it in, for example, a CRM system or similar.
Questions?
If you have questions about the above, you are always welcome to contact support.