diff --git a/docs/speech-usage.rst b/docs/speech-usage.rst
index 5ba566432fdd..3846125a79f9 100644
--- a/docs/speech-usage.rst
+++ b/docs/speech-usage.rst
@@ -88,14 +88,14 @@ Great Britian.
     >>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
     ...                        encoding=speech.Encoding.FLAC,
     ...                        sample_rate=44100)
-    >>> results = sample.sync_recognize(
+    >>> alternatives = sample.sync_recognize(
     ...     speech.Encoding.FLAC, 16000,
     ...     source_uri='gs://my-bucket/recording.flac', language_code='en-GB',
     ...     max_alternatives=2)
-    >>> for result in results:
+    >>> for alternative in alternatives:
     ...     print('=' * 20)
-    ...     print('transcript: ' + result.transcript)
-    ...     print('confidence: ' + result.confidence)
+    ...     print('transcript: ' + alternative.transcript)
+    ...     print('confidence: ' + alternative.confidence)
     ====================
     transcript: Hello, this is a test
     confidence: 0.81
@@ -112,12 +112,12 @@ Example of using the profanity filter.
     >>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
     ...                        encoding=speech.Encoding.FLAC,
     ...                        sample_rate=44100)
-    >>> results = sample.sync_recognize(max_alternatives=1,
-    ...                                 profanity_filter=True)
-    >>> for result in results:
+    >>> alternatives = sample.sync_recognize(max_alternatives=1,
+    ...                                      profanity_filter=True)
+    >>> for alternative in alternatives:
     ...     print('=' * 20)
-    ...     print('transcript: ' + result.transcript)
-    ...     print('confidence: ' + result.confidence)
+    ...     print('transcript: ' + alternative.transcript)
+    ...     print('confidence: ' + alternative.confidence)
     ====================
     transcript: Hello, this is a f****** test
     confidence: 0.81
@@ -134,12 +134,12 @@ words to the vocabulary of the recognizer.
     ...                        encoding=speech.Encoding.FLAC,
     ...                        sample_rate=44100)
     >>> hints = ['hi', 'good afternoon']
-    >>> results = sample.sync_recognize(max_alternatives=2,
-    ...                                 speech_context=hints)
-    >>> for result in results:
+    >>> alternatives = sample.sync_recognize(max_alternatives=2,
+    ...                                      speech_context=hints)
+    >>> for alternative in alternatives:
     ...     print('=' * 20)
-    ...     print('transcript: ' + result.transcript)
-    ...     print('confidence: ' + result.confidence)
+    ...     print('transcript: ' + alternative.transcript)
+    ...     print('confidence: ' + alternative.confidence)
     ====================
     transcript: Hello, this is a test
     confidence: 0.81
diff --git a/speech/google/cloud/speech/_gax.py b/speech/google/cloud/speech/_gax.py
index 2465ec0e50bd..3c343f2227b5 100644
--- a/speech/google/cloud/speech/_gax.py
+++ b/speech/google/cloud/speech/_gax.py
@@ -16,14 +16,12 @@
 
 
 from google.cloud.gapic.speech.v1beta1.speech_client import SpeechClient
-from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import RecognitionAudio
-from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
-    RecognitionConfig)
-from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
-    SpeechContext)
-from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
+from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import RecognitionAudio
+from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import RecognitionConfig
+from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import SpeechContext
+from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
     StreamingRecognitionConfig)
-from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
+from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
     StreamingRecognizeRequest)
 from google.longrunning import operations_grpc
 
@@ -31,8 +29,8 @@
 from google.cloud._helpers import make_secure_stub
 from google.cloud._http import DEFAULT_USER_AGENT
 
+from google.cloud.speech.alternative import Alternative
 from google.cloud.speech.operation import Operation
-from google.cloud.speech.result import Result
 
 OPERATIONS_API_HOST = 'speech.googleapis.com'
 
@@ -237,9 +235,15 @@ def sync_recognize(self, sample, language_code=None, max_alternatives=None,
                                words to the vocabulary of the recognizer.
 
         :rtype: list
-        :returns: List of :class:`google.cloud.speech.result.Result` objects.
+        :returns: A list of dictionaries. One dict for each alternative. Each
+                  dictionary typically contains two keys (though not
+                  all will be present in all cases)
 
-        :raises: ValueError if there are no results.
+                  * ``transcript``: The detected text from the audio recording.
+                  * ``confidence``: The confidence in language detection, float
+                    between 0 and 1.
+
+        :raises: ValueError if more than one result is returned or no results.
         """
         config = RecognitionConfig(
             encoding=sample.encoding, sample_rate=sample.sample_rate,
@@ -250,13 +254,13 @@ def sync_recognize(self, sample, language_code=None, max_alternatives=None,
                                  uri=sample.source_uri)
         api = self._gapic_api
         api_response = api.sync_recognize(config=config, audio=audio)
-
-        # Sanity check: If we got no results back, raise an error.
-        if len(api_response.results) == 0:
-            raise ValueError('No results returned from the Speech API.')
-
-        # Iterate over any results that came back.
-        return [Result.from_pb(result) for result in api_response.results]
+        if len(api_response.results) == 1:
+            results = api_response.results.pop()
+            alternatives = results.alternatives
+            return [Alternative.from_pb(alternative)
+                    for alternative in alternatives]
+        else:
+            raise ValueError('More than one result or none returned from API.')
 
 
 def _stream_requests(sample, language_code=None, max_alternatives=None,
diff --git a/speech/google/cloud/speech/client.py b/speech/google/cloud/speech/client.py
index 0bf96c68e100..f6bcdf50b26a 100644
--- a/speech/google/cloud/speech/client.py
+++ b/speech/google/cloud/speech/client.py
@@ -60,11 +60,7 @@ class Client(BaseClient):
     def __init__(self, credentials=None, http=None, use_gax=None):
         super(Client, self).__init__(credentials=credentials, http=http)
         self._connection = Connection(
-            credentials=self._credentials,
-            http=self._http,
-        )
-
-        # Save on the actual client class whether we use GAX or not.
+            credentials=self._credentials, http=self._http)
         if use_gax is None:
             self._use_gax = _USE_GAX
         else:
diff --git a/speech/google/cloud/speech/operation.py b/speech/google/cloud/speech/operation.py
index 330328ce0b3b..f56f0602fa1c 100644
--- a/speech/google/cloud/speech/operation.py
+++ b/speech/google/cloud/speech/operation.py
@@ -14,10 +14,10 @@
 
 """Long running operation representation for Google Speech API"""
 
-from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
 from google.cloud import operation
-from google.cloud.speech.result import Result
+from google.cloud.speech.alternative import Alternative
 
 
 operation.register_type(cloud_speech_pb2.AsyncRecognizeMetadata)
@@ -58,13 +58,11 @@ def _update_state(self, operation_pb):
         if result_type != 'response':
             return
 
-        # Retrieve the results.
-        # If there were no results at all, raise an exception.
         pb_results = self.response.results
-        if len(pb_results) == 0:
-            raise ValueError('Speech API returned no results.')
+        if len(pb_results) != 1:
+            raise ValueError('Expected exactly one result, found:',
+                             pb_results)
 
-        # Save the results to the Operation object.
-        self.results = []
-        for pb_result in pb_results:
-            self.results.append(Result.from_pb(pb_result))
+        result = pb_results[0]
+        self.results = [Alternative.from_pb(alternative)
+                        for alternative in result.alternatives]
diff --git a/speech/google/cloud/speech/result.py b/speech/google/cloud/speech/result.py
index bba01a047c5d..11efb93626b6 100644
--- a/speech/google/cloud/speech/result.py
+++ b/speech/google/cloud/speech/result.py
@@ -17,53 +17,6 @@
 from google.cloud.speech.alternative import Alternative
 
 
-class Result(object):
-    """Speech recognition result representation.
-
-    This is the object that comes back on sync or async requests
-    (but not streaming requests).
-
-    :type alternatives: list
-    :param alternatives: List of
-        :class:`~google.cloud.speech.alternative.Alternative`.
-    """
-    def __init__(self, alternatives):
-        self.alternatives = alternatives
-
-    @classmethod
-    def from_pb(cls, result):
-        """Factory: construct instance of ``SpeechRecognitionResult``.
-
-        :type result: :class:`~google.cloud.grpc.speech.v1beta1\
-                               .cloud_speech_pb2.StreamingRecognizeResult`
-        :param result: Instance of ``StreamingRecognizeResult`` protobuf.
-
-        :rtype: :class:`~google.cloud.speech.result.SpeechRecognitionResult`
-        :returns: Instance of ``SpeechRecognitionResult``.
-        """
-        alternatives = [Alternative.from_pb(result) for result
-                        in result.alternatives]
-        return cls(alternatives=alternatives)
-
-    @property
-    def confidence(self):
-        """Return the confidence for the most probable alternative.
-
-        :rtype: float
-        :returns: Confidence value, between 0 and 1.
-        """
-        return self.alternatives[0].confidence
-
-    @property
-    def transcript(self):
-        """Return the transcript for the most probable alternative.
-
-        :rtype: str
-        :returns: Speech transcript.
-        """
-        return self.alternatives[0].transcript
-
-
 class StreamingSpeechResult(object):
     """Streaming speech result representation.
 
@@ -93,27 +46,9 @@ def from_pb(cls, response):
         :rtype: :class:`~google.cloud.speech.result.StreamingSpeechResult`
         :returns: Instance of ``StreamingSpeechResult``.
         """
-        alternatives = [Alternative.from_pb(result) for result
-                        in response.alternatives]
+        alternatives = [Alternative.from_pb(alternative)
+                        for alternative in response.alternatives]
         is_final = response.is_final
         stability = response.stability
         return cls(alternatives=alternatives, is_final=is_final,
                    stability=stability)
-
-    @property
-    def confidence(self):
-        """Return the confidence for the most probable alternative.
-
-        :rtype: float
-        :returns: Confidence value, between 0 and 1.
-        """
-        return self.alternatives[0].confidence
-
-    @property
-    def transcript(self):
-        """Return the transcript for the most probable alternative.
-
-        :rtype: str
-        :returns: Speech transcript.
-        """
-        return self.alternatives[0].transcript
diff --git a/speech/setup.py b/speech/setup.py
index aa8796d18b57..eb10b0e7dff2 100644
--- a/speech/setup.py
+++ b/speech/setup.py
@@ -51,7 +51,7 @@
 REQUIREMENTS = [
     'google-cloud-core >= 0.22.1, < 0.23dev',
     'grpcio >= 1.0.2, < 2.0dev',
-    'gapic-google-cloud-speech-v1beta1 >= 0.15.0, < 0.16dev',
+    'gapic-google-cloud-speech-v1beta1 >= 0.14.0, < 0.15dev',
 ]
 
 setup(
diff --git a/speech/unit_tests/test__gax.py b/speech/unit_tests/test__gax.py
index b3e40edc1797..7ee1b0945fa5 100644
--- a/speech/unit_tests/test__gax.py
+++ b/speech/unit_tests/test__gax.py
@@ -36,8 +36,13 @@ def _call_fut(self, sample, language_code, max_alternatives,
     def test_ctor(self):
         from google.cloud import speech
         from google.cloud.speech.sample import Sample
-        from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
-            RecognitionConfig, SpeechContext, StreamingRecognitionConfig,
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
+            SpeechContext)
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
+            RecognitionConfig)
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
+            StreamingRecognitionConfig)
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
             StreamingRecognizeRequest)
 
         sample = Sample(content=self.AUDIO_CONTENT,
@@ -98,8 +103,10 @@ def test_stream_requests(self):
         from io import BytesIO
         from google.cloud import speech
         from google.cloud.speech.sample import Sample
-        from google.cloud.proto.speech.v1beta1.cloud_speech_pb2 import (
-            StreamingRecognitionConfig, StreamingRecognizeRequest)
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
+            StreamingRecognitionConfig)
+        from google.cloud.grpc.speech.v1beta1.cloud_speech_pb2 import (
+            StreamingRecognizeRequest)
 
         sample = Sample(stream=BytesIO(self.AUDIO_CONTENT),
                         encoding=speech.Encoding.FLAC,
diff --git a/speech/unit_tests/test_alternative.py b/speech/unit_tests/test_alternative.py
index 7a79d37e43b8..950392ae517d 100644
--- a/speech/unit_tests/test_alternative.py
+++ b/speech/unit_tests/test_alternative.py
@@ -54,7 +54,7 @@ def test_from_api_repr_with_no_confidence(self):
         self.assertIsNone(alternative.confidence)
 
     def test_from_pb_with_no_confidence(self):
-        from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+        from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
         text = 'the double trouble'
         pb_value = cloud_speech_pb2.SpeechRecognitionAlternative(
diff --git a/speech/unit_tests/test_client.py b/speech/unit_tests/test_client.py
index 43d3527ec339..a5c37d9c3d0a 100644
--- a/speech/unit_tests/test_client.py
+++ b/speech/unit_tests/test_client.py
@@ -24,7 +24,7 @@ def _make_credentials():
 
 
 def _make_result(alternatives=()):
-    from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+    from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
     return cloud_speech_pb2.SpeechRecognitionResult(
         alternatives=[
@@ -37,7 +37,7 @@ def _make_result(alternatives=()):
 
 
 def _make_streaming_result(alternatives=(), is_final=True, stability=1.0):
-    from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+    from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
     return cloud_speech_pb2.StreamingRecognitionResult(
         alternatives=[
@@ -52,7 +52,7 @@ def _make_streaming_result(alternatives=(), is_final=True, stability=1.0):
 
 
 def _make_streaming_response(*results):
-    from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+    from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
     response = cloud_speech_pb2.StreamingRecognizeResponse(
         results=results,
@@ -61,7 +61,7 @@ def _make_streaming_response(*results):
 
 
 def _make_sync_response(*results):
-    from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+    from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
     response = cloud_speech_pb2.SyncRecognizeResponse(
         results=results,
@@ -202,7 +202,7 @@ def test_sync_recognize_source_uri_without_optional_params_no_gax(self):
         sample = client.sample(source_uri=self.AUDIO_SOURCE_URI,
                                encoding=encoding, sample_rate=self.SAMPLE_RATE)
 
-        response = [i for i in sample.sync_recognize()]
+        response = sample.sync_recognize()
 
         self.assertEqual(len(client._connection._requested), 1)
         req = client._connection._requested[0]
@@ -231,7 +231,7 @@ def test_sync_recognize_with_empty_results_no_gax(self):
                                sample_rate=self.SAMPLE_RATE)
 
         with self.assertRaises(ValueError):
-            next(sample.sync_recognize())
+            sample.sync_recognize()
 
     def test_sync_recognize_with_empty_results_gax(self):
         from google.cloud._testing import _Monkey
@@ -274,7 +274,7 @@ def speech_api(channel=None):
                                sample_rate=self.SAMPLE_RATE)
 
         with self.assertRaises(ValueError):
-            next(sample.sync_recognize())
+            sample.sync_recognize()
 
     def test_sync_recognize_with_gax(self):
         from google.cloud._testing import _Monkey
@@ -326,19 +326,16 @@ def speech_api(channel=None):
         self.assertEqual(
             channel_args, [(creds, _gax.DEFAULT_USER_AGENT, host)])
 
-        results = [i for i in sample.sync_recognize()]
+        results = sample.sync_recognize()
 
-        self.assertEqual(len(results), 1)
-        self.assertEqual(len(results[0].alternatives), 2)
+        self.assertEqual(len(results), 2)
         self.assertEqual(results[0].transcript,
-                         results[0].alternatives[0].transcript,
                          alternatives[0]['transcript'])
         self.assertEqual(results[0].confidence,
-                         results[0].alternatives[0].confidence,
                          alternatives[0]['confidence'])
-        self.assertEqual(results[0].alternatives[1].transcript,
+        self.assertEqual(results[1].transcript,
                          alternatives[1]['transcript'])
-        self.assertEqual(results[0].alternatives[1].confidence,
+        self.assertEqual(results[1].confidence,
                          alternatives[1]['confidence'])
 
     def test_async_supported_encodings(self):
@@ -538,11 +535,9 @@ def speech_api(channel=None):
         self.assertEqual(results[0].stability, 0.122435)
         self.assertEqual(results[1].stability, 0.1432343)
         self.assertFalse(results[1].is_final)
-        self.assertEqual(results[1].transcript,
-                         results[1].alternatives[0].transcript,
+        self.assertEqual(results[1].alternatives[0].transcript,
                          alternatives[0]['transcript'])
-        self.assertEqual(results[1].confidence,
-                         results[1].alternatives[0].confidence,
+        self.assertEqual(results[1].alternatives[0].confidence,
                          alternatives[0]['confidence'])
         self.assertEqual(results[1].alternatives[1].transcript,
                          alternatives[1]['transcript'])
@@ -550,11 +545,9 @@ def speech_api(channel=None):
                          alternatives[1]['confidence'])
         self.assertTrue(results[2].is_final)
         self.assertEqual(results[2].stability, 0.9834534)
-        self.assertEqual(results[2].transcript,
-                         results[2].alternatives[0].transcript,
+        self.assertEqual(results[2].alternatives[0].transcript,
                          alternatives[0]['transcript'])
-        self.assertEqual(results[2].confidence,
-                         results[2].alternatives[0].confidence,
+        self.assertEqual(results[2].alternatives[0].confidence,
                          alternatives[0]['confidence'])
 
     def test_stream_recognize(self):
diff --git a/speech/unit_tests/test_operation.py b/speech/unit_tests/test_operation.py
index cbcac1a363a7..8a890d134720 100644
--- a/speech/unit_tests/test_operation.py
+++ b/speech/unit_tests/test_operation.py
@@ -44,7 +44,7 @@ def test_constructor(self):
 
     @staticmethod
     def _make_result(transcript, confidence):
-        from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+        from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
 
         return cloud_speech_pb2.SpeechRecognitionResult(
             alternatives=[
@@ -56,7 +56,7 @@ def _make_result(transcript, confidence):
         )
 
     def _make_operation_pb(self, *results):
-        from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
+        from google.cloud.grpc.speech.v1beta1 import cloud_speech_pb2
         from google.longrunning import operations_pb2
         from google.protobuf.any_pb2 import Any
 
@@ -86,48 +86,31 @@ def test__update_state_no_response(self):
 
     def test__update_state_with_response(self):
         from google.cloud.speech.alternative import Alternative
-        from google.cloud.speech.result import Result
 
         client = object()
         operation = self._make_one(
             self.OPERATION_NAME, client)
 
-        results = [
-            self._make_result('hi mom', 0.75),
-            self._make_result('hi dad', 0.75),
-        ]
-        operation_pb = self._make_operation_pb(*results)
+        text = 'hi mom'
+        confidence = 0.75
+        result = self._make_result(text, confidence)
+        operation_pb = self._make_operation_pb(result)
         operation._update_state(operation_pb)
         self.assertIsNotNone(operation.response)
 
-        self.assertEqual(len(operation.results), 2)
-        for result, text in zip(operation.results, ['hi mom', 'hi dad']):
-            self.assertIsInstance(result, Result)
-            self.assertEqual(result.transcript, text)
-            self.assertEqual(result.confidence, 0.75)
-            self.assertIsInstance(result.alternatives, list)
-            self.assertIsInstance(result.alternatives[0], Alternative)
+        self.assertEqual(len(operation.results), 1)
+        alternative = operation.results[0]
+        self.assertIsInstance(alternative, Alternative)
+        self.assertEqual(alternative.transcript, text)
+        self.assertEqual(alternative.confidence, confidence)
 
-    def test__update_state_with_empty_response(self):
-        from google.cloud.proto.speech.v1beta1 import cloud_speech_pb2
-        from google.longrunning import operations_pb2
-        from google.protobuf.any_pb2 import Any
-
-        # Simulate an empty response (rather than no response yet, which
-        # is distinct).
-        response = cloud_speech_pb2.AsyncRecognizeResponse(results=[])
-        type_url = 'type.googleapis.com/%s' % response.DESCRIPTOR.full_name
-        any_pb = Any(
-            type_url=type_url,
-            value=response.SerializeToString(),
-        )
-        operation_pb = operations_pb2.Operation(
-            name=self.OPERATION_NAME,
-            response=any_pb,
-        )
-
-        # Establish that we raise ValueError at state update time.
+    def test__update_state_bad_response(self):
         client = object()
-        operation = self._make_one(self.OPERATION_NAME, client)
+        operation = self._make_one(
+            self.OPERATION_NAME, client)
+
+        result1 = self._make_result('is this ok?', 0.625)
+        result2 = self._make_result('ease is ok', None)
+        operation_pb = self._make_operation_pb(result1, result2)
         with self.assertRaises(ValueError):
             operation._update_state(operation_pb)