如何解析Google Speech API返回的JSON格式语音识别结果？

Hey there! Let’s walk through how to parse that Google Speech-to-Text response you’ve got stored in your data variable. I’ll use Python for examples since it’s the go-to for this kind of JSON parsing task, but the core logic translates to other languages too.

First, a quick recap: your data variable is a Python dictionary (converted from the JSON response) from the LongRunningRecognize API—this is the endpoint used for longer audio files that require asynchronous processing.

Step 1: Verify the job completed #

First, check if the recognition job finished successfully using the done flag:

if data['done']:
    print("Recognition job completed! Let's parse the results.")
else:
    print("Job is still in progress—check back later.")

Step 2: Extract the core response data #

The actual transcription results are nested inside the response key. Let’s grab that first to simplify access:

recognition_response = data['response']

Step 3: Pull the transcript and confidence score #

The results list holds all transcribed segments (for longer audio, there will be multiple entries here). For your example, there’s one result entry. Let’s extract the top transcript (the highest-confidence option, which is always first in the alternatives list):

# Get the first transcribed segment (adjust the index if you need later segments)
first_segment = recognition_response['results'][0]

# Extract the alternative transcript options
transcript_options = first_segment['alternatives']

# Grab the top transcript and its confidence score
top_transcript = transcript_options[0]['transcript']
confidence_score = transcript_options[0]['confidence']  # Your snippet cut this off, but this is the full key

print(f"Transcript: {top_transcript}")
print(f"Confidence Score: {confidence_score}")

Step 4: Access job metadata (optional) #

If you need details about the job’s timeline or progress, use the metadata key, which follows the LongRunningRecognizeMetadata schema:

job_metadata = data['metadata']
progress = job_metadata['progressPercent']
start_time = job_metadata['startTime']
completion_time = job_metadata['lastUpdateTime']

print(f"Job started at: {start_time}")
print(f"Job finished at: {completion_time}")

Handling longer audio with multiple segments #

For longer audio files, the results list will have multiple entries (each representing a distinct audio segment). You can loop through all segments to collect every transcript:

for segment_num, result in enumerate(recognition_response['results'], start=1):
    best_option = result['alternatives'][0]
    print(f"Segment {segment_num}: {best_option['transcript']} (Confidence: {best_option['confidence']})")

Pro Tips to Avoid Errors: #

• Use dict.get() to safely handle missing keys (e.g., if a transcript doesn’t have a confidence score):

  confidence_score = transcript_options[0].get('confidence', 'No score available')
  

• The alternatives list includes all possible transcriptions sorted by confidence—you can access lower-confidence options if needed, but the first one is almost always the most accurate.

内容的提问来源于stack exchange，提问作者user9316498