Add support for audio file upload

app/api/tests/test_utils.py

@@ -4,9 +4,10 @@ from django.test import TestCase
 
 from seqeval.metrics.sequence_labeling import get_entities
 
+from ..exceptions import FileParseException
 from ..models import Label, Document
 from ..utils import BaseStorage, ClassificationStorage, SequenceLabelingStorage, Seq2seqStorage, CoNLLParser
-from ..utils import iterable_to_io
+from ..utils import AudioParser, iterable_to_io
 
 
 class TestBaseStorage(TestCase):
@@ -138,6 +139,26 @@ class TestCoNLLParser(TestCase):
         })
 
 
+class TestAudioParser(TestCase):
+    def test_parse_mp3(self):
+        f = io.BytesIO(b'...')
+        f.name = 'test.mp3'
+
+        actual = next(AudioParser().parse(f))
+
+        self.assertEqual(actual, [{
+            'audio': 'data:audio/mpeg;base64,Li4u',
+            'meta': '{"filename": "test.mp3"}',
+        }])
+
+    def test_parse_unknown(self):
+        f = io.BytesIO(b'...')
+        f.name = 'unknown.unknown'
+
+        with self.assertRaises(FileParseException):
+            next(AudioParser().parse(f))
+
+
 class TestIterableToIO(TestCase):
     def test(self):
         def iterable():

app/api/utils.py

@@ -1,7 +1,9 @@
+import base64
 import csv
 import io
 import itertools
 import json
+import mimetypes
 import re
 from collections import defaultdict
 
@@ -227,7 +229,9 @@ class Speech2textStorage(BaseStorage):
     @transaction.atomic
     def save(self, user):
         for data in self.data:
-            doc = self.save_doc([{'text': audio['audio']} for audio in data])
+            for audio in data:
+                audio['text'] = audio.pop('audio')
+            doc = self.save_doc(data)
             annotations = self.make_annotations(doc, data)
             self.save_annotation(annotations, user)
 
@@ -411,6 +415,19 @@ class JSONParser(FileParser):
             yield data
 
 
+class AudioParser(FileParser):
+    def parse(self, file):
+        file_type, _ = mimetypes.guess_type(file.name, strict=False)
+        if not file_type:
+            raise FileParseException(line_num=1, line='Unable to guess file type')
+
+        audio = base64.b64encode(file.read())
+        yield [{
+            'audio': f'data:{file_type};base64,{audio.decode("ascii")}',
+            'meta': json.dumps({'filename': file.name}),
+        }]
+
+
 class JSONLRenderer(JSONRenderer):
 
     def render(self, data, accepted_media_type=None, renderer_context=None):

app/api/views.py

@@ -18,7 +18,7 @@ from .models import Project, Label, Document, RoleMapping, Role
 from .permissions import IsProjectAdmin, IsAnnotatorAndReadOnly, IsAnnotator, IsAnnotationApproverAndReadOnly, IsOwnAnnotation, IsAnnotationApprover
 from .serializers import ProjectSerializer, LabelSerializer, DocumentSerializer, UserSerializer
 from .serializers import ProjectPolymorphicSerializer, RoleMappingSerializer, RoleSerializer
-from .utils import CSVParser, ExcelParser, JSONParser, PlainTextParser, CoNLLParser, iterable_to_io
+from .utils import CSVParser, ExcelParser, JSONParser, PlainTextParser, CoNLLParser, AudioParser, iterable_to_io
 from .utils import JSONLRenderer
 from .utils import JSONPainter, CSVPainter
 
@@ -243,6 +243,8 @@ class TextUploadAPI(APIView):
             return CoNLLParser()
         elif file_format == 'excel':
             return ExcelParser()
+        elif file_format == 'audio':
+            return AudioParser()
         else:
             raise ValidationError('format {} is invalid.'.format(file_format))
 

app/server/static/components/annotation.pug

@@ -74,7 +74,9 @@ div.columns(v-cloak="")
         )
           span.icon
             i.fa.fa-check(v-show="annotations[index] && annotations[index].length")
-          span.name {{ doc.text.slice(0, 60) }}...
+          span.name
+            span(v-if="documentMetadataFor(index) && documentMetadataFor(index).filename") {{ documentMetadataFor(index).filename }}
+            span(v-else) {{ doc.text.slice(0, 60) }}...
 
   div.column.is-7.is-offset-1.message.hero.is-fullheight#message-pane
 

app/server/static/components/annotationMixin.js

@@ -177,6 +177,20 @@ export default {
       });
     },
 
+    documentMetadataFor(i) {
+      const document = this.docs[i];
+      if (document == null || document.meta == null) {
+        return null;
+      }
+
+      const metadata = JSON.parse(document.meta);
+      if (isEmpty(metadata)) {
+        return null;
+      }
+
+      return metadata;
+    },
+
     getState() {
       if (this.picked === 'all') {
         return '';
@@ -292,17 +306,7 @@ export default {
     },
 
     documentMetadata() {
-      const document = this.docs[this.pageNumber];
-      if (document == null || document.meta == null) {
-        return null;
-      }
-
-      const metadata = JSON.parse(document.meta);
-      if (isEmpty(metadata)) {
-        return null;
-      }
-
-      return metadata;
+      return this.documentMetadataFor(this.pageNumber);
     },
 
     id2label() {

app/server/static/components/examples/upload_speech2text.jsonl

@@ -1,3 +1,3 @@
-{"audio": "data:audio/mpeg;base64,..."}
+{"audio": "data:audio/mpeg;base64,...", "meta":{"filename": "a-story.mp3"}}
 {"audio": "https://server.com/audio.ogg"}
 {"audio": "data:audio/wav;base64,...", "transcription": "Once upon a time..."}

app/server/static/components/upload_speech2text.vue

@@ -12,6 +12,16 @@ block select-format-area
     )
     | JSONL
 
+  label.radio
+    input(
+      type="radio"
+      name="format"
+      value="audio"
+      v-bind:checked="format === 'audio'"
+      v-model="format"
+    )
+    | Audio
+
 block example-format-area
   pre.code-block(v-show="format === 'json'")
     code.json

app/server/templates/dataset.html

@@ -24,7 +24,7 @@
         {% for doc in object_list %}
         <tr>
           <td>{{ forloop.counter0|add:page_obj.start_index }}</td>
-          <td>{{ doc.text|truncatechars:200 }}</td>
+          <td>{{ doc.text|truncatechars:100 }}</td>
           <td>
             <p class="control">
               <button class="button is-text delete-document-button" data-delete-document-id="{{ doc.id }}">