Add file-like object support to StreamWriter (#2648)

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/2648

Reviewed By: nateanl

Differential Revision: D38976874

Pulled By: mthrok

fbshipit-source-id: 0541dea2a633d97000b4b8609ff6b83f6b82c864

test/torchaudio_unittest/io/stream_writer_test.py

 import torch
 import torchaudio
 
-from parameterized import parameterized
+from parameterized import parameterized, parameterized_class
 from torchaudio_unittest.common_utils import (
     get_asset_path,
     is_ffmpeg_available,
@@ -42,8 +42,42 @@ def get_video_chunk(fmt, frame_rate, *, width, height):
     return chunk
 
 
+################################################################################
+# Helper decorator and Mixin to duplicate the tests for fileobj
+_media_source = parameterized_class(
+    ("test_fileobj",),
+    [(False,), (True,)],
+    class_name_func=lambda cls, _, params: f'{cls.__name__}{"_fileobj" if params["test_fileobj"] else "_path"}',
+)
+
+
+class _MediaSourceMixin:
+    def setUp(self):
+        super().setUp()
+        self.src = None
+
+    def get_dst(self, path):
+        if not self.test_fileobj:
+            return path
+        if self.src is not None:
+            raise ValueError("get_dst can be called only once.")
+
+        self.src = open(path, "wb")
+        return self.src
+
+    def tearDown(self):
+        if self.src is not None:
+            self.src.flush()
+            self.src.close()
+        super().tearDown()
+
+
+################################################################################
+
+
 @skipIfNoFFmpeg
-class StreamWriterInterfaceTest(TempDirMixin, TorchaudioTestCase):
+@_media_source
+class StreamWriterInterfaceTest(_MediaSourceMixin, TempDirMixin, TorchaudioTestCase):
     @classmethod
     def setUpClass(cls):
         super().setUpClass()
@@ -55,7 +89,7 @@ class StreamWriterInterfaceTest(TempDirMixin, TorchaudioTestCase):
         super().tearDownClass()
 
     def get_dst(self, path):
-        return self.get_temp_path(path)
+        return super().get_dst(self.get_temp_path(path))
 
     def get_buf(self, path):
         with open(self.get_temp_path(path), "rb") as fileobj:
@@ -70,8 +104,8 @@ class StreamWriterInterfaceTest(TempDirMixin, TorchaudioTestCase):
         sample_rate = 8000
         num_channels = 1
 
-        path = self.get_dst("test.mp3")
-        s = StreamWriter(path, format="mp3")
+        dst = self.get_dst("test.mp3")
+        s = StreamWriter(dst, format="mp3")
         s.set_metadata(metadata={"artist": "torchaudio", "title": "foo"})
         s.set_metadata(metadata={"title": self.id()})
         s.add_audio_stream(sample_rate, num_channels, format=src_fmt)
@@ -80,6 +114,7 @@ class StreamWriterInterfaceTest(TempDirMixin, TorchaudioTestCase):
         with s.open():
             s.write_audio_chunk(0, chunk)
 
+        path = self.get_temp_path("test.mp3")
         tag = TinyTag.get(path)
         assert tag.artist is None
         assert tag.title == self.id()
@@ -203,6 +238,8 @@ class StreamWriterInterfaceTest(TempDirMixin, TorchaudioTestCase):
             s.write_video_chunk(0, chunk)
 
         # Fetch the written data
+        if self.test_fileobj:
+            dst.flush()
         buf = self.get_buf(filename)
         result = torch.frombuffer(buf, dtype=torch.uint8)
         if encoder_fmt.endswith("p"):

torchaudio/csrc/CMakeLists.txt

@@ -203,6 +203,7 @@ if (BUILD_TORCHAUDIO_PYTHON_EXTENSION)
       ffmpeg/pybind/typedefs.cpp
       ffmpeg/pybind/pybind.cpp
       ffmpeg/pybind/stream_reader.cpp
+      ffmpeg/pybind/stream_writer.cpp
       )
     torchaudio_extension(
       _torchaudio_ffmpeg

torchaudio/csrc/ffmpeg/pybind/pybind.cpp

 #include <pybind11/stl.h>
 #include <torch/extension.h>
 #include <torchaudio/csrc/ffmpeg/pybind/stream_reader.h>
+#include <torchaudio/csrc/ffmpeg/pybind/stream_writer.h>
 
 namespace torchaudio {
 namespace ffmpeg {
 namespace {
 
 PYBIND11_MODULE(_torchaudio_ffmpeg, m) {
+  py::class_<StreamWriterFileObj, c10::intrusive_ptr<StreamWriterFileObj>>(
+      m, "StreamWriterFileObj")
+      .def(py::init<py::object, const c10::optional<std::string>&, int64_t>())
+      .def("set_metadata", &StreamWriterFileObj::set_metadata)
+      .def("add_audio_stream", &StreamWriterFileObj::add_audio_stream)
+      .def("add_video_stream", &StreamWriterFileObj::add_video_stream)
+      .def("dump_format", &StreamWriterFileObj::dump_format)
+      .def("open", &StreamWriterFileObj::open)
+      .def("write_audio_chunk", &StreamWriterFileObj::write_audio_chunk)
+      .def("write_video_chunk", &StreamWriterFileObj::write_video_chunk)
+      .def("flush", &StreamWriterFileObj::flush)
+      .def("close", &StreamWriterFileObj::close);
   py::class_<StreamReaderFileObj, c10::intrusive_ptr<StreamReaderFileObj>>(
       m, "StreamReaderFileObj")
       .def(py::init<
            py::object,
            const c10::optional<std::string>&,
-           const c10::optional<std::map<std::string, std::string>>&,
+           const c10::optional<OptionMap>&,
            int64_t>())
       .def("num_src_streams", &StreamReaderFileObj::num_src_streams)
       .def("num_out_streams", &StreamReaderFileObj::num_out_streams)

torchaudio/csrc/ffmpeg/pybind/stream_reader.cpp

@@ -27,7 +27,7 @@ StreamReaderFileObj::StreamReaderFileObj(
     const c10::optional<std::string>& format,
     const c10::optional<std::map<std::string, std::string>>& option,
     int64_t buffer_size)
-    : FileObj(fileobj_, static_cast<int>(buffer_size)),
+    : FileObj(fileobj_, static_cast<int>(buffer_size), false),
       StreamReaderBinding(get_input_format_context(
           static_cast<std::string>(py::str(fileobj_.attr("__str__")())),
           format,

torchaudio/csrc/ffmpeg/pybind/stream_writer.cpp

+#include <torchaudio/csrc/ffmpeg/pybind/stream_writer.h>
+
+namespace torchaudio {
+namespace ffmpeg {
+
+StreamWriterFileObj::StreamWriterFileObj(
+    py::object fileobj_,
+    const c10::optional<std::string>& format,
+    int64_t buffer_size)
+    : FileObj(fileobj_, static_cast<int>(buffer_size), true),
+      StreamWriterBinding(get_output_format_context(
+          static_cast<std::string>(py::str(fileobj_.attr("__str__")())),
+          format,
+          pAVIO)) {}
+
+void StreamWriterFileObj::set_metadata(
+    const std::map<std::string, std::string>& metadata) {
+  StreamWriter::set_metadata(map2dict(metadata));
+}
+
+void StreamWriterFileObj::add_audio_stream(
+    int64_t sample_rate,
+    int64_t num_channels,
+    std::string format,
+    const c10::optional<std::string>& encoder,
+    const c10::optional<std::map<std::string, std::string>>& encoder_option,
+    const c10::optional<std::string>& encoder_format) {
+  StreamWriter::add_audio_stream(
+      sample_rate,
+      num_channels,
+      format,
+      encoder,
+      map2dict(encoder_option),
+      encoder_format);
+}
+
+void StreamWriterFileObj::add_video_stream(
+    double frame_rate,
+    int64_t width,
+    int64_t height,
+    std::string format,
+    const c10::optional<std::string>& encoder,
+    const c10::optional<std::map<std::string, std::string>>& encoder_option,
+    const c10::optional<std::string>& encoder_format) {
+  StreamWriter::add_video_stream(
+      frame_rate,
+      width,
+      height,
+      format,
+      encoder,
+      map2dict(encoder_option),
+      encoder_format);
+}
+
+} // namespace ffmpeg
+} // namespace torchaudio

torchaudio/csrc/ffmpeg/pybind/stream_writer.h

+#pragma once
+#include <torchaudio/csrc/ffmpeg/pybind/typedefs.h>
+#include <torchaudio/csrc/ffmpeg/stream_writer/stream_writer_wrapper.h>
+
+namespace torchaudio {
+namespace ffmpeg {
+
+class StreamWriterFileObj : protected FileObj, public StreamWriterBinding {
+ public:
+  StreamWriterFileObj(
+      py::object fileobj,
+      const c10::optional<std::string>& format,
+      int64_t buffer_size);
+
+  void set_metadata(const std::map<std::string, std::string>&);
+  void add_audio_stream(
+      int64_t sample_rate,
+      int64_t num_channels,
+      std::string format,
+      const c10::optional<std::string>& encoder,
+      const c10::optional<std::map<std::string, std::string>>& encoder_option,
+      const c10::optional<std::string>& encoder_format);
+  void add_video_stream(
+      double frame_rate,
+      int64_t width,
+      int64_t height,
+      std::string format,
+      const c10::optional<std::string>& encoder,
+      const c10::optional<std::map<std::string, std::string>>& encoder_option,
+      const c10::optional<std::string>& encoder_format);
+};
+
+} // namespace ffmpeg
+} // namespace torchaudio

torchaudio/csrc/ffmpeg/pybind/typedefs.cpp

@@ -31,6 +31,16 @@ static int read_function(void* opaque, uint8_t* buf, int buf_size) {
   return num_read == 0 ? AVERROR_EOF : num_read;
 }
 
+static int write_function(void* opaque, uint8_t* buf, int buf_size) {
+  FileObj* fileobj = static_cast<FileObj*>(opaque);
+  buf_size = FFMIN(buf_size, fileobj->buffer_size);
+
+  py::bytes b(reinterpret_cast<const char*>(buf), buf_size);
+  // TODO: check the return value to check
+  fileobj->fileobj.attr("write")(b);
+  return buf_size;
+}
+
 static int64_t seek_function(void* opaque, int64_t offset, int whence) {
   // We do not know the file size.
   if (whence == AVSEEK_SIZE) {
@@ -40,7 +50,17 @@ static int64_t seek_function(void* opaque, int64_t offset, int whence) {
   return py::cast<int64_t>(fileobj->fileobj.attr("seek")(offset, whence));
 }
 
-AVIOContextPtr get_io_context(FileObj* opaque, int buffer_size) {
+AVIOContextPtr get_io_context(FileObj* opaque, int buffer_size, bool writable) {
+  if (writable) {
+    TORCH_CHECK(
+        py::hasattr(opaque->fileobj, "write"),
+        "`write` method is not available.");
+  } else {
+    TORCH_CHECK(
+        py::hasattr(opaque->fileobj, "read"),
+        "`read` method is not available.");
+  }
+
   uint8_t* buffer = static_cast<uint8_t*>(av_malloc(buffer_size));
   TORCH_CHECK(buffer, "Failed to allocate buffer.");
 
@@ -50,10 +70,10 @@ AVIOContextPtr get_io_context(FileObj* opaque, int buffer_size) {
   AVIOContext* av_io_ctx = avio_alloc_context(
       buffer,
       buffer_size,
-      0,
+      writable ? 1 : 0,
       static_cast<void*>(opaque),
       &read_function,
-      nullptr,
+      writable ? &write_function : nullptr,
       py::hasattr(opaque->fileobj, "seek") ? &seek_function : nullptr);
 
   if (!av_io_ctx) {
@@ -64,25 +84,28 @@ AVIOContextPtr get_io_context(FileObj* opaque, int buffer_size) {
 }
 } // namespace
 
-FileObj::FileObj(py::object fileobj_, int buffer_size)
+FileObj::FileObj(py::object fileobj_, int buffer_size, bool writable)
     : fileobj(fileobj_),
       buffer_size(buffer_size),
-      pAVIO(get_io_context(this, buffer_size)) {}
+      pAVIO(get_io_context(this, buffer_size, writable)) {}
 
-c10::optional<OptionDict> map2dict(
-    const c10::optional<std::map<std::string, std::string>>& src) {
-  if (!src) {
-    return {};
-  }
+OptionDict map2dict(const OptionMap& src) {
   OptionDict dict;
-  for (const auto& it : src.value()) {
+  for (const auto& it : src) {
     dict.insert(it.first.c_str(), it.second.c_str());
   }
-  return c10::optional<OptionDict>{dict};
+  return dict;
+}
+
+c10::optional<OptionDict> map2dict(const c10::optional<OptionMap>& src) {
+  if (src) {
+    return c10::optional<OptionDict>{map2dict(src.value())};
+  }
+  return {};
 }
 
-std::map<std::string, std::string> dict2map(const OptionDict& src) {
-  std::map<std::string, std::string> ret;
+OptionMap dict2map(const OptionDict& src) {
+  OptionMap ret;
   for (const auto& it : src) {
     ret.insert({it.key(), it.value()});
   }

torchaudio/csrc/ffmpeg/pybind/typedefs.h

@@ -9,13 +9,16 @@ struct FileObj {
   py::object fileobj;
   int buffer_size;
   AVIOContextPtr pAVIO;
-  FileObj(py::object fileobj, int buffer_size);
+  FileObj(py::object fileobj, int buffer_size, bool writable);
 };
 
-c10::optional<OptionDict> map2dict(
-    const c10::optional<std::map<std::string, std::string>>& src);
+using OptionMap = std::map<std::string, std::string>;
 
-std::map<std::string, std::string> dict2map(const OptionDict& src);
+OptionDict map2dict(const OptionMap& src);
+
+c10::optional<OptionDict> map2dict(const c10::optional<OptionMap>& src);
+
+OptionMap dict2map(const OptionDict& src);
 
 } // namespace ffmpeg
 } // namespace torchaudio

torchaudio/csrc/ffmpeg/stream_writer/stream_writer.cpp

@@ -485,7 +485,8 @@ void StreamWriter::open(const c10::optional<OptionDict>& option) {
   // file-like object)
   AVFORMAT_CONST AVOutputFormat* fmt = pFormatContext->oformat;
   AVDictionary* opt = get_option_dict(option);
-  if (!(fmt->flags & AVFMT_NOFILE)) {
+  if (!(fmt->flags & AVFMT_NOFILE) &&
+      !(pFormatContext->flags & AVFMT_FLAG_CUSTOM_IO)) {
     ret = avio_open2(
         &pFormatContext->pb,
         pFormatContext->url,
@@ -524,7 +525,8 @@ void StreamWriter::close() {
   // Close the file if it was not provided by client code (i.e. when not
   // file-like object)
   AVFORMAT_CONST AVOutputFormat* fmt = pFormatContext->oformat;
-  if (!(fmt->flags & AVFMT_NOFILE)) {
+  if (!(fmt->flags & AVFMT_NOFILE) &&
+      !(pFormatContext->flags & AVFMT_FLAG_CUSTOM_IO)) {
     // avio_closep can be only applied to AVIOContext opened by avio_open
     avio_closep(&(pFormatContext->pb));
   }

torchaudio/csrc/ffmpeg/stream_writer/stream_writer_wrapper.cpp

@@ -5,7 +5,14 @@ namespace ffmpeg {
 
 AVFormatOutputContextPtr get_output_format_context(
     const std::string& dst,
-    const c10::optional<std::string>& format) {
+    const c10::optional<std::string>& format,
+    AVIOContext* io_ctx) {
+  if (io_ctx) {
+    TORCH_CHECK(
+        format,
+        "`format` must be provided when the input is file-like object.");
+  }
+
   AVFormatContext* p = avformat_alloc_context();
   TORCH_CHECK(p, "Failed to allocate AVFormatContext.");
 
@@ -19,6 +26,11 @@ AVFormatOutputContextPtr get_output_format_context(
       av_err2string(ret),
       ").");
 
+  if (io_ctx) {
+    p->pb = io_ctx;
+    p->flags |= AVFMT_FLAG_CUSTOM_IO;
+  }
+
   return AVFormatOutputContextPtr(p);
 }
 

torchaudio/csrc/ffmpeg/stream_writer/stream_writer_wrapper.h

@@ -7,7 +7,8 @@ namespace ffmpeg {
 // create format context for writing media
 AVFormatOutputContextPtr get_output_format_context(
     const std::string& dst,
-    const c10::optional<std::string>& format);
+    const c10::optional<std::string>& format,
+    AVIOContext* io_ctx = nullptr);
 
 class StreamWriterBinding : public StreamWriter,
                             public torch::CustomClassHolder {

torchaudio/io/_stream_writer.py

 from typing import Dict, Optional
 
 import torch
+import torchaudio
 
 
 def _format_doc(**kwargs):
@@ -50,7 +51,16 @@ class StreamWriter:
 
     Args:
         dst (str): The destination where the encoded data are written.
-            The supported value depends on the FFmpeg found in the system.
+            If string-type, it must be a resource indicator that FFmpeg can
+            handle. The supported value depends on the FFmpeg found in the system.
+
+            If file-like object, it must support `write` method with the signature
+            `write(data: bytes) -> int`.
+
+            Please refer to the following for the expected signature and behavior of
+            `write` method.
+
+            - https://docs.python.org/3/library/io.html#io.BufferedIOBase.write
 
         format (str or None, optional):
             Override the output format, or specify the output media device.
@@ -81,14 +91,25 @@ class StreamWriter:
                https://ffmpeg.org/ffmpeg-devices.html#Output-Devices
 
                Use `ffmpeg -devices` to list the values available in the current environment.
+
+        buffer_size (int):
+            The internal buffer size in byte. Used only when `dst` is a file-like object.
+
+            Default: `4096`.
     """
 
     def __init__(
         self,
         dst: str,
         format: Optional[str] = None,
+        buffer_size: int = 4096,
     ):
-        self._s = torch.classes.torchaudio.ffmpeg_StreamWriter(dst, format)
+        if isinstance(dst, str):
+            self._s = torch.classes.torchaudio.ffmpeg_StreamWriter(dst, format)
+        elif hasattr(dst, "write"):
+            self._s = torchaudio._torchaudio_ffmpeg.StreamWriterFileObj(dst, format, buffer_size)
+        else:
+            raise ValueError("`dst` must be either a string or a file-like object.")
         self._is_open = False
 
     @_format_common_args