Is there a way to get a more complete word alternatives list with watson speech-to-text?

In watson speech-to-text’s word alternatives, I’ve noticed that there are sometimes words that are missing from the word alternatives despite being present in the alternate transcriptions. Is there a way to have the service return word alternatives that include every word that appears in the alt transcriptions?

For example, in this result, the word “and” is present in the last alt transcription, but never appears in the word alternatives:

> {
> “result_index”: 0,
> “results”: [
> {
> “final”: true,
> “alternatives”: [
> {
> “confidence”: 0.609,
> “transcript”: “come parent her “,
> },
> {
> “transcript”: “compounding her ”
> },
> {
> “transcript”: “come pound her ”
> },
> {
> “transcript”: “compound her ”
> },
> {
> “transcript”: “empowering her ”
> },
> {
> “transcript”: “cum parent her ”
> },
> {
> “transcript”: “some parent her ”
> },
> {
> “transcript”: “come power and her ”
> }
> ],
> “word_alternatives”: [
> {
> “start_time”: 0.03,
> “alternatives”: [
> {
> “confidence”: 0.5538,
> “word”: “come”
> },
> {
> “confidence”: 0.1764,
> “word”: “compounding”
> },
> {
> “confidence”: 0.1706,
> “word”: “compound”
> },
> {
> “confidence”: 0.0341,
> “word”: “empowering”
> },
> {
> “confidence”: 0.0255,
> “word”: “cum”
> },
> {
> “confidence”: 0.0196,
> “word”: “some”
> },
> {
> “confidence”: 0.0102,
> “word”: “can”
> },
> {
> “confidence”: 0.0068,
> “word”: “compounded”
> },
> {
> “confidence”: 0.0031,
> “word”: “empowerment”
> }
> ],
> “end_time”: 0.22
> },
> {
> “start_time”: 0.22,
> “alternatives”: [
> {
> “confidence”: 0.394,
> “word”: “parent”
> },
> {
> “confidence”: 0.2076,
> “word”: “pound”
> },
> {
> “confidence”: 0.0051,
> “word”: “power”
> },
> {
> “confidence”: 0.0012,
> “word”: “apparent”
> },
> {
> “confidence”: 0.001,
> “word”: “parents”
> }
> ],
> “end_time”: 0.66
> },
> {
> “start_time”: 0.66,
> “alternatives”: [
> {
> “confidence”: 0.9999,
> “word”: “her”
> }
> ],
> “end_time”: 1.05
> }
> ]
> }
> ]
> }

I’ve also noticed that there are other transcriptions (“compound her” and “empowering her” in the above example) that do not include a word from the second set of word alternatives, but the second set of word alternatives does not include any indication that an alt transcription might not include any of the words. Is there a way to have the service indicate that an alternative might choose not to include any of the word alternatives?

I’m using the java api.

Related:

Leave a Reply