Initial sketch.

asr/__init__.py

@@ -0,0 +1,105 @@
+from email.mime.multipart import MIMEMultipart
+from email.message import Message
+import json
+import struct
+
+from flask import Flask, request, Response
+from google.cloud import speech
+
+
+app = Flask(__name__)
+
+
+# We know gunicorn does this, but it doesn't *say* it does this, so we must signal it manually.
+@app.before_request
+def handle_chunking():
+    request.environ['wsgi.input_terminated'] = 1
+
+
+def parse_chunks(stream):
+    boundary = b'--' + request.headers['content-type'].split(';')[1].split('=')[1].encode('utf-8').strip()  # super lazy/brittle parsing.
+    print("Boundary: " + boundary.decode('utf-8'))
+    this_frame = b''
+    while True:
+        content = stream.read(4096)
+        this_frame += content
+        end = this_frame.find(boundary)
+        if end > -1:
+            frame = this_frame[:end]
+            if frame != b'':
+                header, content = frame.split(b'\r\n\r\n', 1)
+                print(content)
+                yield content[:-2]
+            this_frame = this_frame[end + len(boundary):]
+        if content == b'':
+            print("End of input.")
+            break
+
+
+def parse_data():
+    boundary = b'--' + request.headers['content-type'].split(';')[1].split('=')[1].encode('utf-8').strip()  # super lazy/brittle parsing.
+    parts = request.data.split(boundary)
+    for part in parts:
+        if part == b'':
+            continue
+        yield part.split(b'\r\n\r\n', 1)[1][:-2]
+
+
+@app.route('/NmspServlet/', methods=["POST"])
+def recognise():
+
+    client = speech.SpeechClient()
+    stream = request.stream
+    chunks = iter(list(parse_chunks(stream)))
+    content = next(chunks).decode('utf-8')
+    print(content)
+
+    config = speech.types.RecognitionConfig(
+        encoding='SPEEX_WITH_HEADER_BYTE',
+        language_code='en-US',
+        sample_rate_hertz=16000,
+    )
+    print('beginning request')
+    responses = client.streaming_recognize(
+        config=speech.types.StreamingRecognitionConfig(config=config),
+        requests=(
+            speech.types.StreamingRecognizeRequest(audio_content=struct.pack('B', len(x)) + x)
+            for x in chunks))
+    print('finished request')
+    words = []
+    for response in responses:
+        if response.results:
+            for result in response.results:
+                words.extend({
+                                 'word': x,
+                                 'confidence': result.alternatives[0].confidence
+                             } for x in result.alternatives[0].transcript.split(' '))
+
+    # Now for some reason we also need to give back a mime/multipart message...
+    parts = MIMEMultipart()
+    response_part = Message()
+    response_part.add_header('Content-Type', 'application/JSON; charset=utf-8')
+
+    if len(words) > 0:
+        response_part.add_header('Content-Disposition', 'form-data; name="QueryResult"')
+        words[0]['word'] += '\\*no-space-before'
+        response_part.set_payload(json.dumps({
+            'words': [words],
+        }))
+    else:
+        response_part.add_header('Content-Disposition', 'form-data; name="QueryRetry"')
+        # Other errors probably exist, but I don't know what they are.
+        # This is a Nuance error verbatim.
+        response_part.set_payload(json.dumps({
+            "Cause": 1,
+            "Name": "AUDIO_INFO",
+            "Prompt": "Sorry, speech not recognized. Please try again."
+        }))
+    parts.attach(response_part)
+    print(parts.as_string())
+
+    response = Response(parts.as_string().split("\n", 3)[3])
+    response.headers['Content-Type'] = f'multipart/form-data; boundary={parts.get_boundary()}'
+    response.headers['Connection'] = 'close'
+    return response
+