Hack: add audio transcription

Dockerfile

@@ -96,7 +96,8 @@ RUN apt-get -qq -y update \
     tesseract-ocr-aze \
     tesseract-ocr-bel \
     tesseract-ocr-uzb \
-    ### pdf convert: libreoffice + a bunch of fonts 
+    ffmpeg \
+    ### pdf convert: libreoffice + a bunch of fonts
     libreoffice fonts-opensymbol hyphen-fr hyphen-de \
     hyphen-en-us hyphen-it hyphen-ru fonts-dejavu fonts-dejavu-core fonts-dejavu-extra \
     fonts-droid-fallback fonts-dustin fonts-f500 fonts-fanwood fonts-freefont-ttf \
@@ -144,6 +145,9 @@ RUN python3 -m spacy download el_core_news_sm \
     && python3 -m spacy download da_core_news_sm
 # RUN python3 -m spacy download zh_core_web_sm
 
+RUN pip3 install --no-cache-dir openai-whisper
+RUN whisper --model base /dev/null || true
+
 COPY . /ingestors
 WORKDIR /ingestors
 RUN pip3 install --no-cache-dir -e /ingestors

Makefile

@@ -3,7 +3,6 @@ COMPOSE=docker-compose
 DOCKER=$(COMPOSE) run --rm ingest-file
 
 .PHONY: build
-
 all: build shell
 
 build:
@@ -36,7 +35,11 @@ format-check:
 	black --check .
 
 test: services
-	$(DOCKER) pytest --cov=ingestors --cov-report html --cov-report term
+	$(DOCKER) pytest --cov=ingestors --cov-report html --cov-report term tests/test_doc.py
+
+minialeph:
+	$(COMPOSE) up -d --remove-orphans redis
+	FTM_STORE_URI=hack.db $(COMPOSE) run --rm mini-ingest-file /bin/bash
 
 restart: build
 	$(COMPOSE) up --force-recreate --no-deps --detach ingest-file

docker-compose.yml

@@ -33,3 +33,25 @@ services:
     depends_on:
       - postgres
       - redis
+
+  mini-ingest-file:
+    build:
+      context: .
+    #image: ghcr.io/alephdata/ingest-file
+    hostname: ingest
+    tmpfs:
+      - /tmp:mode=777
+      - /data:mode=777
+    environment:
+      FTM_STORE_URI: sqlite:///hack.db
+      LOG_FORMAT: TEXT  # TEXT or JSON
+    volumes:
+      - "./ingestors:/ingestors/ingestors"
+      - "./tests:/ingestors/tests"
+      - "./data:/ingestors/data"
+      - "./requirements.txt:/ingestors/requirements.txt"
+      - "./setup.py:/ingestors/setup.py"
+      - "./hack.db:/ingestors/hack.db"
+      - "~:/host"
+    depends_on:
+      - redis

ingestors/media/audio.py

@@ -1,6 +1,7 @@
 import logging
 from followthemoney import model
 from pymediainfo import MediaInfo
+import whisper
 
 from ingestors.ingestor import Ingestor
 from ingestors.support.timestamp import TimestampSupport
@@ -41,6 +42,9 @@ def ingest(self, file_path, entity):
         try:
             entity.schema = model.get("Audio")
             metadata = MediaInfo.parse(file_path)
+            whisper_model = whisper.load_model("base")
+            result = whisper_model.transcribe(audio=str(file_path))
+            entity.add("bodyText", result["text"])
             for track in metadata.tracks:
                 entity.add("title", track.title)
                 entity.add("generator", track.writing_application)